Spring 2026 GRASP SFI: Jay Patrikar, Field AI, “Exploiting Uncertainty for Fieldable Robot Autonomy”
March 18 at 3:00 PM - 4:00 PM
This will be a hybrid event with in-person attendance in Amy Gutmann Hall 414 and virtual attendance on Zoom.
ABSTRACT
Capability without deployability is a demo. Deployability without safety is a hazard. Safety without capability is a parked robot. From autonomous driving to humanoids, the challenges remain the same: large-scale internet-trained policies bootstrap impressive capabilities but break down at the edges of deployment. This talk distills lessons from bridging research and field deployment across robot morphologies and domains. The central thesis is that uncertainty, when properly quantified, becomes an actionable signal in the autonomy stack: it tells you when your learned planner is operating outside its training distribution, how it can be used to obtain a semantically grounded notion of safety, and how video world models can become faithful, deployable simulators. The throughline is a shift from optimizing worst-case benchmarks to managing reasonable risk in the field.
Jay Patrikar
Field AI
Jay Patrikar is a Senior Research Scientist at FieldAI, a robotics company building risk-aware foundation models for robots where his work focuses on the safe and trustworthy deployment of autonomous agents in real-world safety-critical domains. During his PhD at Carnegie Mellon University, Jay collaborated with frontier research labs at Microsoft Research and NVIDIA Research. He holds a PhD and MS in Robotics from Carnegie Mellon University’s School of Computer Science and dual degrees (Master and Bachelor of Technology in Aerospace Engineering) from the Indian Institute of Technology Kanpur. Jay’s contributions have been featured in prominent news outlets such as Nature and Popular Science along with receiving the Best Paper Award at AIAA 2024. Additionally, Jay holds an FAA Part 107 Remote Pilot Certificate and an FAA Part 61 Private Pilot Certificate.