Projects

"How to make robots more safe?"

..........The main motive of my research is to find an answer to this question. My research primarily centers on provably safe decision making algorithms specifically in robotic systems. More specifically, I am interested in determining the potential failure modes within robotic algorithms as they engage with their surroundings through a variety of multimodal sensors, including but not limited to images, point clouds, haptics, and language. This knowledge can then serve as a means for enhancing the underlying algorithms, ultimately yielding robust and safe systems.

Safety and Recovery Algorithms for System Level Autonomous Vehicle Perception Failures
Related works: [ArXiv '24]

We propose methods to enhance the safe operation of autonomous vehicles (AVs) in dynamic environments, particularly when sensor failures occur. Our approach incorporates feedback from a runtime monitor to mitigate the impact of missed obstacles on the AV's planning and control. It acts as a safety filter and plan repair system using a lightweight transformer-based model, designed to operate efficiently in real time.

Failure Mitigation of Visual Navigation Systems
Related works: [ArXiv '24][ICRA '24] [Allerton 23] [RA-L, IROS '23]

We propose methods to ensure the safety of robotic systems with machine learning-driven controllers using visual inputs. We identify and mitigate potential system-level failures through innovative techniques like reachability analysis and anomaly detection. By systematically analyzing the behavior of vision-based controllers, we enhance system safety and reliability in real-world applications, particularly in unpredictable environments.


Navigation Using Contact Mechanics
Related works: [RA-L, IROS '22] [APS '22]

We present a new approach for multi-legged robots to utilize leg-obstacle collisions to generate desired dynamics. We demonstrate in experiments that using this method an open-loop quadrupedal robot was able to achieve desired orientations within a periodic obstacle field without any sensory input or active steering.