Active Inference Robotics Skill (Second-Order)

"The agent's job is to predict its actions by predicting its sensations." — Patrick Kenny

Trigger Conditions

• User asks about bridging active inference with robot control
• Questions about predictive coding in locomotion policies
• Connecting KL divergence minimization to RL training
• Mean field approximation in robotics state estimation
• Sim2Real as inference about future observations

Overview

Second-order skill synthesizing Patrick Kenny's discrete active inference framework with K-Scale's JAX/MuJoCo robotics stack. This skill emerges from the constructive collision between:

1. Active Inference Institute (ActInf ModelStream 019.1, Jan 2025)
2. K-Scale Labs (ksim, kos, kinfer ecosystem)
3. MuJoCo Playground (DeepMind's sim2real framework)