Paarth Shah
Papers
2
Total Citations
7
H-Index
2
About
Paarth Shah is an emerging researcher at the intersection of robot learning, imitation learning, and trustworthy autonomy. His work addresses some of the most pressing practical challenges in deploying learned robotic systems in real-world settings — namely, how to evaluate and monitor robot policies when data and testing resources are scarce. His most notable contribution, "Can We Detect Failures Without Failure Data?" (2025), tackles the critical problem of runtime failure detection for imitation learning policies, proposing uncertainty-aware methods that function even in the absence of labeled failure examples — a significant step toward safer robot deployment. This work has already garnered 5 citations within its first year, signaling strong early interest from the robotics community. Complementing this, his 2024 paper on statistically rigorous performance evaluation of behavior cloning policies offers a principled framework for assessing policy generalizability under constrained real-world testing conditions, a challenge widely faced by practitioners working with stochastic, generative visuomotor policies. Together, Shah's research establishes a coherent vision: making robot learning systems not only capable, but reliably evaluable and trustworthy — qualities essential for the responsible real-world deployment of next-generation robotic manipulation systems.
Research Focus
Key Achievements
Top Papers
- 1
- 2