Active Inference as the Test-Time Scaling Law for Physical AI Agents

2026-06-21

AI Summary

This work addresses the limited generalization of physical AI agents in out-of-distribution, unknown environments, where existing scaling laws struggle to cope with the non-stationarity of real-world settings. Drawing on first principles of active inference, the paper proposes a test-time scaling mechanism that dynamically updates both policy and world model via variational inference to minimize expected free energy, thereby enabling continual learning and reasoning during deployment. This approach formalizes active inference within a test-time scaling framework—departing from conventional reliance on model scale and static training data—and uncovers its neural correlates in the basal ganglia and prefrontal cortex. Evaluated in autonomous driving simulations, the method demonstrates substantially improved robust generalization in unseen scenarios and achieves over 36% higher inference efficiency compared to model-free Q-learning and Bayesian reinforcement learning baselines.

Abstract

In this paper, a novel test-time scaling law for physical artificial intelligence (AI) agents is introduced. This scaling law enables physical AI agents to reason with their world models to generalize in unforeseen scenarios at test time. The derived scaling law is grounded in the first principle of active inference, which equips agents with the general objective to survive in the real world, under which their specific task objectives are subsumed. Active inference achieves this by providing the reasoning to resolve prediction errors that arise when the agent encounters unforeseen situations outside its training distribution, enabling generalization in non-stationary environments. The proposed scaling law captures this by dynamically updating the agent's policy with this reasoning at test time. This policy update is modeled as a soft Bayesian inference process in which beliefs about the policy are updated using the reasoning that reduces expected prediction errors under allowable policies as a likelihood. The resulting posterior policy admits a biological interpretation, recovering the scaling mechanism that engages the brain's basal ganglia and prefrontal cortex at test time. To solve this analytically intractable problem, a variational inference solution minimizing free energy bounds is developed. This solution extends to enable learning beyond training by reinforcing new instances, resolved at test time, in both the policy and world model. Unlike existing scaling laws constrained by model size and training data, the derived solution scales with the continuous real-world experience of a physical AI agent. Simulation results on an autonomous driving task demonstrate that the proposed solution outperforms model-free Q-learning and model-based Bayesian reinforcement learning, achieving robust generalization to unforeseen scenarios while improving inference efficiency by over 36%.

Problem

• test-time generalization
• physical AI agents
• unforeseen scenarios
• non-stationary environments
• scaling law

Innovation

• active inference
• test-time scaling law
• physical AI agents
• variational inference
• world model