Abstract
The present disclosure provides a method, a system, and a computer program product for delegating autonomy to sub-agents in a multi-agent system. The method comprises receiving a task at a supervisor agent and obtaining a risk score of the task, an agent success rate of a sub-agent for a task type of the task, and an escalation cost of involving a human reviewer. The method includes computing a continuous trust score using an actor network trained via an actor-critic reinforcement learning method. The method includes determining, based on the trust score and decision thresholds, whether the sub-agent auto-executes the task, executes the task under assisted oversight, or escalates the task to a human reviewer. The method also includes computing, after an outcome of the task is known, a reward based on task success, the escalation cost, and a risk-adjusted failure penalty, and updating the actor network based on the reward.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Mishra, Soumendra Kumar, "SYSTEM AND METHOD FOR AUTONOMY DELEGATION IN MULTI-AGENT SYSTEMS USING REINFORCEMENT LEARNING", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12815