This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
Failure Prediction at Runtime for Generative Robot Policies
Created by
Haebom
Author
Ralf R omer, Adrian Kobras, Luca Worbis, Angela P. Schoellig
Outline
FIPER is a general framework for runtime failure prediction of imitation learning (IL) policies using generative models. It does not require failure data and predicts failure based on out-of-distribution (OOD) observations in the policy embedding space and the high uncertainty of generated actions. FIPER uses conformal prediction to calibrate two metrics and triggers a failure alarm when the aggregated metrics exceed a threshold over a short time window. It has been shown to predict failures more accurately and earlier than existing methods in both simulations and real-world environments.
Takeaways, Limitations
•
Takeaways:
◦
We present a general framework for predicting failures at runtime without using failure data.
◦
We present two key failure indicators leveraging OOD observation and action uncertainty.
◦
We calibrated the failure prediction score using conformal prediction.
◦
It showed improved performance over existing methods in various environments.
◦
Contributed to the development of more interpretable and safe generative robot policies.
•
Limitations:
◦
The specific Limitations was not explicitly mentioned in the paper.
