Aptiv is working towards a realistic “Virtual Driving Simulation” to enhance the verification of Advanced Driver Assistance Systems (ADAS) in various scenarios, including corner cases. Students on the Aptiv team will work with existing real-world vehicle data and develop models to replicate sensor (radar and camera) performance in various environmental conditions, providing realistic sensor outputs that drive the behavior of downstream algorithms.
Aptiv is a global technology company that is developing devices that rewrite the rules of what’s possible to make transportation safer, greener, and more connected. Advanced Driver Assistance Systems (ADAS) sensors, such as radars and cameras, are critical for autonomous vehicles to navigate safely. The real world is full of obstacles, with everyday interferences such as reflective surfaces, whiteout conditions, fog, rain, traffic congestion, and objects such as pedestrians, parked vehicles, buildings, and signs that clutter the environment and can result in diminished sensor performance.
Sensor perception varies in environmental conditions such as snow, fog, and rain (for the camera) and tunnels, barriers, and bridges (for the radar), and verification of ADAS algorithms in these various real-world driving scenarios to achieve robustness is neither a cost-effective solution nor practical in terms of safety.
Students on the Aptiv team will work with existing real-world vehicle data and develop models to replicate sensor (radar and camera) performance in various environmental conditions. These models would provide realistic sensor outputs that drive the behavior of downstream algorithms. This project allows students to work on large data sets and statistical methodologies to characterize radar and camera sensor data under various systemic/environmental conditions in building Level 2+ (semi-automated driving) systems.
Aptiv’s Simulation and Tools department within Aptiv’s Advanced Safety & User Experience division is developing realistic “Virtual Driving Simulation” products to support its internal and external customers. This department believes sensor characterization-based noise injection will help create realistic sensor models in the simulation environment to effectively verify the algorithms across multiple corner cases. Variation in sensor perception impacts downstream software components and vehicle features, and it is imperative to ensure that sensors in the simulation environment are modeled, factoring in external noise factors. Aptiv now aims to develop an innovative “Sensor-Noise Injection” solution that can use their voluminous database of real-time vehicle data to characterize sensors under various external noise factors and environmental conditions.
Minimum Viable Product Deliverable (Minimum level of success)
- The student team will create separate bins for vehicle data logs (provided by Aptiv) based on the environmental noise factors and study how each parameter of the sensor output statistically varies under each noise factor. Larger sample size is recommended for better statistical analysis (do not have to exceed 30 samples).
- Initial mathematical models and equations for each of the parameters of the sensor outputs for use in the “Sensor-Noise Injection” software component to inject noise and create a realistic output in the virtual simulation.
- Verification that mathematical models/equations derive the parametric variations of the sensor output with less than or equal to 5% variation from the sensor output of the vehicle data in a specific environmental condition.
Expected Final Deliverable (Expected level of success)
- A refined version of the model that addresses areas for improvement identified in initial model validation.
Stretch Goal Opportunities: (High level of success)
- Data analysis and model updates for combinational noise factors that simultaneously impact both camera and radar. For example, when analyzing vehicle data where there is heavy rain and an overhead bridge, or dense fog and a tunnel.
Below are the skills needed for this project. Students with the following relevant skills and interest, regardless of major, are encouraged to apply! This is a team based multidisciplinary project. Students on the team are not expected to have experience in all areas, but should be willing to learn and will be asked to perform a breadth of tasks throughout the two semester project.
Data Analysis (2-3 Students)
Specific Skills: Data analysis, modeling, data mining and machine learning techniques, data set cleaning and integration.
Likely Majors: DATA, CS, SI, STATS
Programming Skills (2-3 Students)
Specific Skills: Scripting Languages (Python) in data mining and data analysis
EECS 281 (or equivalent) is required
Likely Majors: DATA, CS
Automotive Knowledge (1-2 Students)
Specific Skills: ADAS, Sensors
EECS 281 (or equivalent) is required
Likely Majors: DATA, ME, EE
Additional Desired Skills/Knowledge/Experience
- Team-based project experience/good team player
- Ability and willingness to teach yourself new skills and techniques
- Big Data modeling experience
- Experience with Python
- Experience or interest in the automotive industry and/or vehicle safety
- Automotive systems knowledge
- In addition to the key technical skills students bring to the team, students will also engage deeply in self-directed learning of new and essential concepts, demonstrate flexibility, collaborate, cooperate, and develop strong professional communication skills.
- Students will need to be able to work outside of their traditional area of study in keeping with the multidisciplinary nature of our projects.
- Aptiv expects professional, respectful behavior always.
Dr. Smitha Vempaty
Dr.Vempaty is a Product Owner for Virtual Driving Simulation at Aptiv. She and her team support internal and external customers to meticulously verify ADAS algorithms with cutting-edge technology, thus increasing the software quality with a cost-effective solution. Smitha has 18 years of automotive experience and holds a Ph.D. in Mechanical Engineering from the University of Ontario Institute of Technology, Canada.
Kerby Shedden is Professor of Statistics in the College of Literature, Science, and the Arts (LSA) and holds a courtesy appointment as Professor of Biostatistics in the School of Public Health. He received a Ph.D. in Statistics from UCLA in 1999. His research focuses on developing and evaluating methods for analyzing high dimensional and complex data including mediation analysis and dimension reduction regression, as well as developing statistical software for applied statistics. He has served as the director of the Consulting for Statistics, Computing, and Analytics Research unit since 2011, and has extensive collaborations with researchers in human biology, nephrology, sleep and aging, and cancer research.
Weekly Meetings: During the winter 2024 semester, the Aptiv team will meet in 277 West Hall on Fridays from 2:00 – 4:00 PM.
Work Location: Most of the work will take place on campus in Ann Arbor. The team will have the opportunity to visit Aptiv’s Technical Center in Troy, MI, for occasional meetings and presentations with stakeholders. MDP will provide transportation.
Course Substitutions: CE MDE, ChE Elective, CS Capstone/MDE, DS Capstone, EE MDE, CoE Honors, IOE Senior Design, ROB 490, ROB 590, SI Elective/Cognate
Citizenship Requirements: This project is open to all students. Note: International students on an F-1 visa will be required to declare part time CPT during Winter 2024 and Fall 2024 terms.
IP/NDA: Students will sign standard University of Michigan IP/NDA documents.
Summer Project Activities: No summer activity will take place on the project.