{"id":5392,"date":"2022-09-08T08:10:52","date_gmt":"2022-09-08T07:10:52","guid":{"rendered":"https:\/\/newminds.se\/?page_id=5392"},"modified":"2025-02-14T14:05:05","modified_gmt":"2025-02-14T14:05:05","slug":"vcc-sva-thesis","status":"publish","type":"page","link":"https:\/\/newminds.se\/eng\/vcc-sva-thesis\/","title":{"rendered":"VCC SVA – Master Thesis 2022"},"content":{"rendered":"
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Volvo Cars - Safe Vehicle Automation
Master's Thesis Proposals 2022<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Read through all of the theses, and then fill out the form at the end of the page. If you already have a master’s thesis partner, it’s only necessary for one of you to submit the info. Rank your top 3 master’s theses subjects that you are interested in. Make sure to do this before Tuesday the 11th of October, 23:59. <\/p>\n

Some of the theses offered have two options, please specify in the form which option is of interest.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Thesis 1<\/strong>: Simulation and Measurements of Recycled Plastics in Front of Radar<\/a><\/p>

Thesis\u00a0<\/strong>2<\/strong>: Simulation for Severity Assessment of Motor Vehicle Accidents<\/a><\/p>

Thesis\u00a0<\/strong>3<\/strong>: Create more Realistic Sensor Models for Virtual Environment, Based on Measured Sensor Performance<\/a><\/p>

Thesis\u00a0<\/strong>4<\/strong>: Modular Architecture for Traffic Controllers in esmini<\/a><\/p>

Thesis\u00a0<\/strong>5<\/strong>: Injection of Simulated Images\/Video to a Real World ECU<\/a><\/p>

Thesis\u00a0<\/strong>6<\/strong>: AI Based Probe Sourcing to Reduce False Positive Warnings in ADAS Equipped Vehicles<\/a><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Thesis\u00a0<\/strong>7<\/strong>: AI Based Road States Estimation for Safe and Energy Efficient Autonomous Driving<\/a><\/p>

Thesis\u00a0<\/strong>8<\/strong>: Distributed Machine Learning Algorithms for Shortest Path Problems with Stochastic Weights <\/a>or Statistical Prediction of Extreme Events to Determine AWD Motor Engagement Strategy<\/a><\/p>

Thesis\u00a0<\/strong>9<\/strong>: Boundary Detection (Upper\/Lower Edge) <\/a>or Extrinsic Lidar Calibration Investigation in Factory Flow<\/a><\/p>

Thesis\u00a0<\/strong>10<\/strong>: Visual Data Analysis of Aggregated Scenarios for Scenario-based Testing of AD Functionality <\/a>or Data-driven Scenario Generation\/Synthesization and Analysis<\/a><\/p>

Thesis\u00a0<\/strong>11<\/strong>: Develop an Ideal Ultrasonic Sensor Model <\/a>or Estimations of Lidar Point Clouds Based on Ultrasonic Sensors Using Deep Learning<\/a><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
\n\t\t\t\t\t\n\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\tfill out the form<\/span>\n\t\t\t\t\t<\/span>\n\t\t\t\t\t<\/a>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis 1<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Simulation and Measurements of Recycled Plastics in Front of Radar<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background:<\/strong> Volvo Cars is committed to reduce our environmental impact. Therefore, besides our goal to become fully electric by 2030, the use of recycled plastics for exterior parts is investigated. Modern vehicles are nowadays equipped with several radar sensors installed behind homogeneous plastic materials, e.g bumpers. Electrical properties of homogenous plastic materials can be measured and their impact on the radar performance can be both measured and simulated. However, recycled plastic present discontinuities in the electrical proprieties which might affect the radar performance in a different way compared to the homogenous material.<\/p>

Objective:<\/strong> Investigate the performance of radar behind recycled plastics<\/p>

Scope:<\/strong>
1. Create method for measuring electrical properties for recycled plastics – (Measurements will be done in RF and Microwave lab at Chalmers. The student will also be responsible for building the measurement setup with the supervision of both Chalmers and Volvo Cars. It might be required to design a scanner to scan the beam over a plastic sample.)<\/p>

2. Create model and method for simulating radar performance behind recycled plastics – (Data analysis of the measurements will need to be performed to extract the electrical properties and it might also be necessary to come up with a new modelling technique to extract such properties. The electrical properties of recycled plastic will then be imported into a simulation tool for simulation and evaluation of their impact on the radar performance.)<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Communication engineering or Wireless photonics and space engineering background, radio frequency labs during education<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Python, Mathlab, ANSYS, CST<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>2<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Simulation for Severity Assessment of Motor Vehicle Accidents<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background:\u00a0<\/strong>Volvo Cars are among the world\u2019s safest cars. We have been leading the safety in automotive industry for decades and contributed in developing new safety features to the road vehicle to protect our customers. At Longitudinal Vehicle Control we work with all the acceleration and deceleration functions with the focus on creating the safest, most energy efficient and rewarding longitudinal vehicle behaviour. The longitudinal control software has a high level of safety criticality (functional safety classification), with acceleration capability around +- 1g. To know how to design the safest software it is required for us to understand how severe the potential traffic hazards can be hence leading to correct identification and classification. We would like to build a simple but also general model to help our design in early concept phase for our software safety activities.<\/p>

Scope:\u00a0<\/strong>This master thesis includes identifying and investigating the most common longitudinal traffic hazard scenarios then build the mass point simulation model based on Simulink. Necessary literature study is required in order to understand and set up the scope of the hazard scenario as well as the safety analysis process from ISO 26262. Testing and verification of the model will be an optional task depending on the time schedule.<\/p>

The thesis work will include the following parts:
\u2022 Literature review across the previous research both within and outside of Volvo Cars
\u2022 Literature study on the ISO 26262: Road vehicles \u2013 Functional safety
\u2022 Identify the potential hazards in longitudinal traffic and set up the research scope
\u2022 Build mass-point based model in Simulink
\u2022 Data post processing to convert the result to severity level defined in ISO 26262
\u2022 Potential optional task for improving: design testing event accordingly and try to get controllability level<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Vehicle dynamics and control, Mechanical engineering, or Mechanical engineering<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Simulink is meritious, as well as a driver\u2019s license<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>3<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Create more Realistic Sensor Models for Virtual Environment, Based on Measured Sensor Performance<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background:<\/strong>\u00a0Sensor models in our virtual environments are ideal, this will make them more non-ideal.\u00a0Simulations in the in-house developed virtual environment (CSPAS) are based on ideal sensors which may lead to more lenient function triggers. In order to get a more realistic sensor detection in the low fidelity sensor models, there is potential to improve the detection algorithm in ideal models by using real sensor measurement data from test cars.<\/span><\/p>

Objectives:<\/strong>\u00a0More realistic sensor performance in virtual environments<\/p>

Scope:<\/strong>\u00a0
1. Select a candidate sensor property for modelling.
2. Develop a method for applying that sensor property.
3. Create model of the measured performance of that property<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Data science, Computer science, Engineering physics, or Mathematics<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

C++<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>4<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Modular Architecture for Traffic Controllers in esmini<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background:\u00a0<\/strong>In OpenSCENARIO there are multiple TrafficActions that will spawn and create random traffic to challenge an AD car. To create different behaviors of the traffic, different Controllers can be used, distributed between the different entities on the road.<\/p>

Objective:\u00a0<\/strong>In order to include traffic in a nice way in esmini (https:\/\/github.com\/esmini\/esmini), a modular architecture is needed to add different driver behaviors. Eg. of modules: Lane change, change speed, overtake, \u201cright hand rule\u201d, road network awareness (like traffic rules).<\/p>

Scope:<\/strong>
1. A modular architecture is created
2. Some example modules integrated<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Computer science<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

C++<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>5<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Injection of Simulated Images\/Video to a Real World ECU<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background:\u00a0<\/strong>Our development centers heavily around software (SW) and in our next generation vehicles, we take a giant leap into the future by launching a capable centralized compute platform, hosting most of the active safety, driver support and vehicle motion control SW. Connecting all exterior sensors to central computer(s) and embedded GPUs is a great enabler for in-house SW development, continuous innovation on a stable compute platform, and superior customer offerings. Together with our partners we are building a modular, service-based perception platform and leading safety and driver support applications. We are setting the scene for future safety innovations as well as autonomous driving features. We deliver on our purpose of providing safe, personal, and sustainable mobility aiming towards a future of Zero Collisions.

In our next generation vehicles, we are building a computer-in-the-car architecture, a key for enabling innovation within areas such as Vehicle Motion, Advanced Connectivity, Machine Learning, and Autonomous Drive. Our mission is to create a Vehicle Control Unit platform using new technologies like DriveOS and NVIDIA\u2019s latest chip technology in combination with more traditional car signaling technologies. This core computer and the corresponding platform will simplify the software\/application development to create a safe, reliable, and secure platform for increased innovation and speed.

Scope:\u00a0<\/strong>Verification and Validation of Advanced Driver Assistance Systems (ADAS) and Autonomous Driving Systems (ADS) with the extensive sensor technology are one of the main challenges to achieve secure and safe self-driving cars. Hardware-in-the-Loop (HIL) based testing methods offer the great advantage of verifying sensors and validating components and systems in an early stage of the development cycle. Advanced sensor simulation techniques can be used to validate functions for autonomous driving throughout the development process.<\/p>

Hardware-in-Loop test environments secure the possibility to test and verify the real ECUs in the controlled and repeatable test environment by stimulating them with recorded data in open loop or synthetic data in closed loop. Consider an example setup for closed-loop HIL simulation as well as raw data input for a front camera. Here, the camera image sensor, including the lens, is replaced and simulated by the HIL environment.<\/p>

The traffic, driver and vehicle dynamics simulation is executed on a Real-Time Platform with an update interval of x msec. In addition, the Real-Time Platform for the restbus simulation is connected to the vehicle network (CAN, Ethernet, FlexRay, etc.). (Restbus simulation is a technique used to validate ECU functions by simulating parts of an in-vehicle bus such as the controller area network.)
The results of this simulation are transferred to a powerful computer which is referred to as the Sensor Simulation PC. The computer then generates a three-dimensional, real-time, high resolution representation of the environment. Using the relevant, parameterized camera model, raw sensor data (camera images in this example) are generated in this PC. The raw sensor data (camera images) is transferred to the Interface Unit via the Display Port interface of the Sensor Simulation PC. This FPGA-based platform executes all remaining parts of the camera model. For example, it executes light control or simulates the I2C interface of the image sensor. This Interface Unit converts the images into electrical signals fed through fiber-optic cable (over GMSL2 protocol) to the Image Processing stack in the Automotive Embedded ECU.

Objectives:<\/strong>\u00a0To evaluate and implement the injection of the simulated (raw) image data directly into the ECU, by research and application in several areas as mentioned below:
1. Since the optical path generated from the camera, consisting of a lens and the imager, is removed; an emulation of these components is necessary in the visualization.
2. The I2C data also needs to be embedded
3. The other aspects related to timing and embedding of data that does not contain the actual image information need to be located and can thus be adapted and parametrized depending on the camera used. Whether an FPGA or a Microcontroller is best suited for this objective, and how they should be programmed to achieve this, is the main research scope in this thesis. Partial Emulation or Complete Emulation of 1) and 2) can be done on the GPU and partially on Microcontroller\/FPGA based Interface Unit, leading to the second research area. For 3), it is evaluated to be best embedded on the Interface Unit, but that is another research scope in this project.<\/p>

4. The connection between Sensor Simulation PC and the Interface Unit is facilitated via the Display Port output or any other viable interface. It is necessary to enable low-latency and efficient transmission of image data including reliability on the image quality with precise timing. This high-precision timing enables limited data buffering resulting in minimizing the lag caused by signal adjustment. Selection of this interface is another research aspect.
5. Evaluation of this prototype system shall be carried out for raw camera data injection at an approximate rate of 10 Gbit\/sec and the possibility to support multiple channels of data streams.<\/p>

As an extension of this thesis work, following research areas are also possible
6. Potentially, multiple camera views can be simulated together in the Sensor Simulation PC, for e.g., a combination of conventional wide-angle and fish-eye cameras. These views need to be optimally synchronized with each other to be able to feed the ECU and the Image Processing stack with accurate timing for the Image Processing output to be viable for Object Detection.
7. The quality of Sensor Fusion\/Object Detection Algorithms shall depend on the accuracy of images being injected. A sensitivity analysis can be carried out by adding pixel errors on the Sensor Simulation PC or fault insertion on the FPGA.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

System controls and mechatronics, or Electrical engineering background. Proactive and communicative, motivated by challenging tasks, teamwork, sharing knowledge, problem solver<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Knowledge about FPGA vs microcontroller programming differences, communication protocols (ethernet, GMSL2, I2C), modeling and simulation.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>6<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

AI Based Probe Sourcing to Reduce False Positive Warnings in ADAS Equipped Vehicles<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background:<\/strong>\u00a0In Connected Safety we receive friction data to be processed in the cloud. We use this friction data to determine if we should generate a slippery road alert. Some of these alerts has proven to be false positives and they seem to be consistently over the same road segment which have speed bumps or similar features.<\/p>

Could there be a hidden pattern that makes it possible to detect and filter out these parts of the road when generating slippery road alerts using machine learning?<\/p>

Objectives:<\/strong>\u00a0Devise a method to process the stream of friction measurements in a way that makes it possible to detect a pattern and identify anomalies, for e.g. speed bumps, to reduce false positive warnings.<\/p>

Scope:<\/strong>\u00a0<\/p>

1. Devise a method to process a stream of measurements as input into a training data set
2. Build a model that can be used in our production environment
3. Investigate the effectiveness of detecting false positives vs real slippery road
4. Investigate what other road features we might be able to detect<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Machine learning, signal processing, control engineering, Java, other programming\/scripting languages for your own analysis and algorithm development<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>7<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

AI Based Road States Estimation for Safe and Energy Efficient Autonomous Driving<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background: <\/strong>Autonomous driving and Electrification are among the major focus areas for Volvo cars. To aid this, next generation platform is equipped with increasingly advanced environment perception sensors. It becomes necessary that algorithms running in the cars should harvest the full potential of advanced sensors to reduce energy consumption and ensure safe autonomous driving. For sending optimal control inputs to brake and propulsion actuators for energy efficiency and safety, knowing conditions and states of the road are critical for decision making algorithms in the car. In this thesis we address the possibility to detect the surface of the road and estimate different properties with the help of deep learning, sensor fusion, image processing and\/or signal processing algorithms.<\/p>

Scope: <\/strong>The thesis work will provide an effective algorithm to estimate the road surface profile in front of the ego vehicle using environment perception sensors on board. The algorithm aims to show the possibility to extract some of the road states such as road inclination, road banks, road condition, bumps, cracks, potholes which can be interesting for vehicle controls and energy optimisation functions. Sensor data which are of interests are lidar, camera, IMU (accelerometer & gyroscope) from next generation electric vehicle platform. Robustness of the algorithm should be tested offline on simple scenarios for example driving at constant speed on straight roads in simulation tool. The scope could also be extended to run the algorithm online on a Nvidia hardware in a real test vehicle.<\/p>

The thesis work will include the following parts:<\/p>

\u2022 Literature review across the state-of-art road detection, estimation and 3D reconstruction algorithms based on lidar and camera data.
\u2022 Fast and efficient transformation of road surface from global coordinates frame to sensor frames of reference.
\u2022 Evaluate the effectiveness of widely used grip-map based estimation of road profiles.
\u2022 Lidar is the important sensor that will be explored in this thesis, so the expectation is to use, for example particle filters for road profile estimation.
\u2022 Camera data will be investigated using computer vision based deep learning and\/or image processing algorithms.
\u2022 Open loop simulation and verification in simulation environment for development and evaluation of the algorithms.
\u2022 Stretched target: Implement the designed algorithm from the thesis on a Nvidia hardware connected to the sensors in a real test vehicle.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

C++ or Python, machine learning, signal processing, vehicle dynamics, control theory<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>8<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Option 1: Distributed Machine Learning Algorithms for Shortest Path Problems with Stochastic Weights<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background<\/strong>: <\/strong>Shortest path problems are present in many different domains, such as in-car navigation systems or computer network routing. When dealing with real-world networks, the costs (e.g., travel time or energy consumption) associated with links in the network may be stochastic instead of fixed. If the cost distributions are known, it is possible to solve the shortest path algorithm using a standard shortest path algorithm with respect to, e.g., the mean costs of all links in the network. When they are not initially known, however, it may be desirable to continuously learn the distribution parameters by observing the costs incurred from tried paths.<\/p>

To use the resources available for this in an efficient way, the problem can be modelled as a sequential decision-making problem. Specifically, one can cast this as a combinatorial version of the classical multi-armed bandit problem. In order to collect enough data to find paths expected to minimize incurred costs, some exploration is often required. On the other hand, too much exploration incurs excessive costs. This tradeoff has previously been addressed through variants of well-known bandit algorithms like Thompson Sampling and UCB, adapted for online shortest path problems with both empirical and theoretical success.<\/p>

However, these methods typically depend on applying standard shortest path algorithms on networks with link weights sequentially modified in various ways to encourage exploration. While they work well for small or moderately sized networks, they are not viable for large networks. Existing shortest path algorithms for deterministic real-world settings (e.g., navigation) may perform concurrent computations in a distributed way across many networked computers to achieve greater performance at scale.\u00a0<\/p>

Objective<\/strong>: <\/strong>The purpose of this project is to combine distributed shortest path methods with online machine learning algorithms and use them on large-scale graphs with stochastic weights.<\/p>

Scope<\/strong>: <\/strong>Adapt combinatorial online learning methods, like e.g., Thompson Sampling or UCB, to a distributed \/ cloud computing setting.<\/p>

Investigate possibility of locating different parts of large road network graphs on separate network nodes, distributing use of computational resources and memory resources accordingly.<\/p>

Perform experimental study to evaluate performance of developed algorithms in terms of run-time efficiency and cumulative regret (sum of path weights incurred over time).<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

MSc. In Engineering physics, Mathematics, Systems & control, or similar. Preferebly with some courses in: mathematics, statistics, physics, control and optimization, machine learning and data mining<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>8<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Option 2: Statistical Prediction of Extreme Events to Determine AWD Motor Engagement Strategy<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background<\/strong>: <\/strong>For a battery electric vehicle (BEV) there is a generally a trade-off between performance and energy efficiency. To obtain maximum performance for an all-wheel drive (AWD) powertrain, both the front and the rear axle motor must be engaged. However, the energy consumption can be decreased by disengaging one of the motors when the power demand is low and AWD is not needed to ensure vehicle stability. Nevertheless, this comes with a cost in terms of performance since it will take some time to re-engage the motor once it is needed.<\/p>

Objective<\/strong>: <\/strong>The objective of the thesis is to estimate the statistical occurrence of extreme events for a specific vehicle and driver, e.g. in terms of requested power demand and vehicle skidding. A powertrain control strategy for motor engagement should then be derived based on the statistical model and a statistical performance requirement, e.g. such that only a specific X-percentile extreme event will lead to a loss of performance.\u00a0<\/p>

Scope<\/strong>: <\/strong>Develop a recursive methodology to estimate a statistical model for extreme events, e.g. based on block maxima and generalized extreme value distributions.<\/span><\/p>

Develop a motor engagement control strategy for an AWD BEV, based on the statistical model for the extreme events.<\/p>

A simulation study where the developed motor engagement strategy is compared to the nominal motor engagement strategy.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Data science, Computer science, Engineering physics, or Mathematics. At least one of the pair needs to be from Chalmers, since the project is a part of a PhD research project in collaboration with Chalmers<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>9<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Option 1: Boundary Detection (Upper\/Lower Edge)<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

This will contribute to our service calibration, to make it possible to create a fixture with a hole in front of the car to use to simplify the problem of detecting the edge of the glass.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Matlab or Python, CAD, modeling, sensor calibration, optimization algorithms, computer vision<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>9<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Option 2: Extrinsic Lidar Calibration Investigation in Factory Flow<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

This kicks off the investigation towards dynamic extrinsic calibration by using nominal values as a base for calibration instead of using factory calibration.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Matlab, Jmodeling, Unity, knowledge of localisation algorithm<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>10<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Option 1: Visual Data Analysis of Aggregated Scenarios for Scenario-based Testing of AD Functionality<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background:<\/strong> Safety assessment of autonomous driving systems are performed using multiple approaches such as Scenario-based testing, Formal verification, Function-based, Shadow mode, Real-world testing, Staged introduction. The Scenario-based approach is a promising method where diverse traffic situations are simulated to test the Autonomouos Driving functionality. Knowledge-based and Data-driven are two predominant approaches to extract scenarios from real-world driving data.<\/p>

Objectives<\/strong>:<\/strong> Scenarios such as cut-ins, cut-outs that are extracted from real-world driving exhibits variations due to, neighbourhood traffic, geographical locations, velocity of the objects and other interesting parameters. Visual analysis methods can help us to understand these variations and identfiy interesting patterns for specific scenario types under different external conditions.<\/p>

Scope<\/strong>: <\/strong><\/p>

1. Literature study to identify the class of visual analysis methods to study scenario data that exhibits spatial and temporal characteristics.
2. Visualize data analysis to uncover patterns in scenarios extracted using knowledge-based and data-driven methods.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Python, spatio-temporal data analysis, dashboarding tools such as Bokeh, Plotly, or Matplotlib<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>10<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Option 2: Data-driven Scenario Generation\/Synthesization and Analysis<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Continuation of previous work in https:\/\/arxiv.org\/abs\/2007.14524<\/a><\/p>

Background:<\/strong> In order to assure safety in self driving cars, AD functionality needs to pass safety tests not only based on \u201creal\u201d scenarios (also called test cases) collected from field driving tests, but also based on many perturbed (similar) trajectories that might have not been collected in real driving data collection. We need to synthesize those while assuring that the generated distributions respects the original data distribution. In previous work, we have implemented and investigated performance of GAN variants to generate scenario trajectories and assessing their quality (w.r.t. the original real data).<\/span><\/p>

Scope & Objectives<\/strong>: In this work, we focus on:
1. Enhancement of the implemented scenario trajectory generation\/synthesization in https:\/\/arxiv.org\/abs\/2007.14524
2. More towards deployment of the already implemented algorithms.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Computer science, complex adaptive systems<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Python, development skills, experience in devops and MLOP is meritious, machine learning, deep learning<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>11<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Option 1: Develop an Ideal Ultrasonic Sensor Model<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background:\u00a0<\/strong>Using simulations in 3D environments to test autonomous functions is being increasingly used as an alternative to real world testing. This has the benefits of being able to test ideas earlier in development as well as testing more complex and dangerous scenarios than is feasible on a test track. But while there are sensor simulations for camera, radar as well as lidar, we currently lack the ability to get corresponding simulated sensor data from ultrasonic sensors. These are commonly used for various parking functions.<\/p>

Scope:\u00a0<\/strong>The aim of this master\u2019s thesis is to develop an ultrasonic sensor model that can be used to simulate data in a 3D environment such as Unity. The performance of the sensor should mimic as much as possible the sensor behavior seen in real world data collection.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Computer science<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

C# or C++, game engines<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Thesis <\/span>11<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Option 2: Estimations of Lidar Point Clouds Based on Ultrasonic Sensors Using Deep Learning<\/h3>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Background:\u00a0<\/strong>In order to achieve full autonomy, vehicles need to accurately perceive their environment. The area closest to the vehicle is usually covered by ultrasonic sensors which has the benefit of a lower cost and smaller size compared to other sensors, but lack the detail given by, for example, a Lidar.

When validating sensor data, one or more Lidars are usually used as a reference sensors, to provide a ground truth against which the production intended sensors are compared. We would like to use this ground truth, not just for validation, but also to investigate if it can be used to train a neural network to achieve a more accurate environmental perception than what is possible with currently existing traditional algorithms.<\/p>

Scope:<\/strong>
The aim of this Master Thesis is to train a neural network which uses Ultrasonic sensor data as input and generates point clouds that mimic the point clouds generated by the more detailed short range reference system Lidars.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Profile<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Computer science<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t

\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
Skills<\/h5>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Python or C++, sensor knowledge, machine learning<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n
\n

<\/p>

    <\/ul><\/div>\n
    \n
    \n<\/fieldset>\n