Research:
I spent my doctoral career to develop feasible solutions that enable autonomous driving in mixed-autonomy environments, where autonomous agents can interact with each other and with humans. Broadly, my research can be divided into:
- Coordination among autonomous agents in the presence of humans,
- Understanding and modeling the behavior of human drivers,
- Enabling reliable and efficient communication between vehicles in the large-scale.
- For a full list of my publications:
Selected Research Projects:
We tackle the problem of autonomous driving in complex & competitive mixed-autonomy environments where autonomous vehicles interact with vehicles driven by humans. Formally, we model this problem as a partially-observable stochastic game and train reinforcement learning agents that cooperate with each other and sympathize with human-driven vehicles. Our autonomous agents learn general latent representations that enables them to coordinate in novel environments.
Studying the data recorded from vehicles driven by humans reveals that a human driver’s behavior consists of certain patterns and micro-maneuvers. Extracting abstract models of these micro-maneuvers enables us to reliably predict the motion and dynamics of a human-driven vehicles. We particularly focus on two approaches: 1) a stochastic hybrid system based on Gaussian and Dirichlet processes and 2) a data-driven model trained on our D2CAV driving dataset.
Vehicle-to-vehicle (V2V) communication enables human-driven (HV) and autonomous vehicles (AVs) to share their situational awareness and constitute a form of mass intelligence to overcome the limitations of a single agent planning in a decentralized fashion. Specifically, we proposed methods that enable the 3GPP C-V2X communication technology to handle thousands of vehicles in heavily congested environments. I presented one of the earliest performance analysis & proposed perhaps the first congestion control algorithm implementation for C-V2X. The outcomes of my work contributed to the development of SAE J3161 standard.
In the first year of my PhD, I led a team of 8 undergraduate and 2 Masters level students to design and build a test platform for research on connected and autonomous vehicles (CAVs). We built a fleet of 1/5 & 1/10 scale vehicles equipped with LiDAR, stereo camera, and an on-board NVIDIA Jetson TX2 computer. The SCVP vehicles are able to reach and maneuver at 60mph speed, making them a realistic testbed for CAV applications. I was responsible for mechanical & electrical design and implementation, developing the autonomous driving software stack, and mentoring the team.
Behrad Toghi Research