C-STEP: Continuous Space-Time Empowerment for Physics-informed Safe Reinforcement Learning of Mobile Agents

2026-03-25

AI Summary

This work proposes C-STEP, an agent-centric safety metric tailored for deterministic continuous domains, to address the challenge of safe navigation for mobile robots in complex environments. C-STEP uniquely integrates continuous space-time empowerment with physics-informed modeling by constructing an interpretable intrinsic reward function that combines the agent’s internal state with forward dynamics. This intrinsic reward is jointly optimized alongside the extrinsic navigation objective, enabling an effective trade-off between task completion and collision avoidance. Experimental results demonstrate that the proposed approach significantly reduces both collision frequency and proximity to obstacles, with only a marginal increase in path duration, thereby validating its efficacy and practicality for safe continuous control in real-world navigation scenarios.

Technology Category

Intelligent Robots: Motion and Path PlanningNatural Language Processing: Safety and RobustnessMultiagent Systems: Adversarial Agents

Application Category

Responsible Web: Machine-in-the-loop, human agency and autonomySecurity and Privacy: Large-scale security measurementsSearch and Retrieval-Augmented AI: Agentic search

Abstract

Safe navigation in complex environments remains a central challenge for reinforcement learning (RL) in robotics. This paper introduces Continuous Space-Time Empowerment for Physics-informed (C-STEP) safe RL, a novel measure of agent-centric safety tailored to deterministic, continuous domains. This measure can be used to design physics-informed intrinsic rewards by augmenting positive navigation reward functions. The reward incorporates the agents internal states (e.g., initial velocity) and forward dynamics to differentiate safe from risky behavior. By integrating C-STEP with navigation rewards, we obtain an intrinsic reward function that jointly optimizes task completion and collision avoidance. Numerical results demonstrate fewer collisions, reduced proximity to obstacles, and only marginal increases in travel time. Overall, C-STEP offers an interpretable, physics-informed approach to reward shaping in RL, contributing to safety for agentic mobile robotic systems.

Problem

• safe navigation
• reinforcement learning
• collision avoidance
• mobile agents
• physics-informed safety

Innovation

• Physics-informed reinforcement learning
• Safe reinforcement learning
• Continuous space-time empowerment
• Intrinsic reward shaping
• Mobile robotic navigation