Predicting Future Actions of Reinforcement Learning Agents

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chung, Stephen, Niekum, Scott, Krueger, David
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917822300422144
author Chung, Stephen
Niekum, Scott
Krueger, David
author_facet Chung, Stephen
Niekum, Scott
Krueger, David
contents As reinforcement learning agents become increasingly deployed in real-world scenarios, predicting future agent actions and events during deployment is important for facilitating better human-agent interaction and preventing catastrophic outcomes. This paper experimentally evaluates and compares the effectiveness of future action and event prediction for three types of RL agents: explicitly planning, implicitly planning, and non-planning. We employ two approaches: the inner state approach, which involves predicting based on the inner computations of the agents (e.g., plans or neuron activations), and a simulation-based approach, which involves unrolling the agent in a learned world model. Our results show that the plans of explicitly planning agents are significantly more informative for prediction than the neuron activations of the other types. Furthermore, using internal plans proves more robust to model quality compared to simulation-based approaches when predicting actions, while the results for event prediction are more mixed. These findings highlight the benefits of leveraging inner states and simulations to predict future agent actions and events, thereby improving interaction and safety in real-world deployments.
format Preprint
id arxiv_https___arxiv_org_abs_2410_22459
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Predicting Future Actions of Reinforcement Learning Agents
Chung, Stephen
Niekum, Scott
Krueger, David
Artificial Intelligence
I.2.6; I.2.8; I.5.1
As reinforcement learning agents become increasingly deployed in real-world scenarios, predicting future agent actions and events during deployment is important for facilitating better human-agent interaction and preventing catastrophic outcomes. This paper experimentally evaluates and compares the effectiveness of future action and event prediction for three types of RL agents: explicitly planning, implicitly planning, and non-planning. We employ two approaches: the inner state approach, which involves predicting based on the inner computations of the agents (e.g., plans or neuron activations), and a simulation-based approach, which involves unrolling the agent in a learned world model. Our results show that the plans of explicitly planning agents are significantly more informative for prediction than the neuron activations of the other types. Furthermore, using internal plans proves more robust to model quality compared to simulation-based approaches when predicting actions, while the results for event prediction are more mixed. These findings highlight the benefits of leveraging inner states and simulations to predict future agent actions and events, thereby improving interaction and safety in real-world deployments.
title Predicting Future Actions of Reinforcement Learning Agents
topic Artificial Intelligence
I.2.6; I.2.8; I.5.1
url https://arxiv.org/abs/2410.22459