This website requires JavaScript.

Online Update of Safety Assurances Using Confidence-Based Predictions

Kensuke NakamuraSomil Bansal
Oct 2022
Robots such as autonomous vehicles and assistive manipulators areincreasingly operating in dynamic environments and close physical proximity topeople. In such scenarios, the robot can leverage a human motion predictor topredict their future states and plan safe and efficient trajectories. However,no model is ever perfect -- when the observed human behavior deviates from themodel predictions, the robot might plan unsafe maneuvers. Recent works haveexplored maintaining a confidence parameter in the human model to overcome thischallenge, wherein the predicted human actions are tempered online based on thelikelihood of the observed human action under the prediction model. This hasopened up a new research challenge, i.e., \textit{how to compute the futurehuman states online as the confidence parameter changes?} In this work, wepropose a Hamilton-Jacobi (HJ) reachability-based approach to overcome thischallenge. Treating the confidence parameter as a virtual state in the system,we compute a parameter-conditioned forward reachable tube (FRT) that providesthe future human states as a function of the confidence parameter. Online, asthe confidence parameter changes, we can simply query the corresponding FRT,and use it to update the robot plan. Computing parameter-conditioned FRTcorresponds to an (offline) high-dimensional reachability problem, which wesolve by leveraging recent advances in data-driven reachability analysis.Overall, our framework enables online maintenance and updates of safetyassurances in human-robot interaction scenarios, even when the human predictionmodel is incorrect. We demonstrate our approach in several safety-criticalautonomous driving scenarios, involving a state-of-the-art deep learning-basedprediction model.