https://masteryip.github.io/tags/humanoid-control/Humanoid ControlHugo Blox Builder (https://hugoblox.com)en-usMon, 21 Sep 2026 00:00:00 +0000https://masteryip.github.io/media/icon_hu7729264130191091259.pnghttps://masteryip.github.io/tags/humanoid-control/https://masteryip.github.io/publication/predactor/Mon, 21 Sep 2026 00:00:00 +0000https://masteryip.github.io/publication/predactor/<h2 id="from-motion-generation-to-direct-control">From motion generation to direct control</h2> <p>PredActor addresses the gap between flexible motion generation and closed-loop execution. A separate reference tracker is replaced by a policy that generates actions directly and uses its own future-state predictions as a guidance interface. Proprioceptive feedback closes the loop as inference rolls forward.</p> <h2 id="steering-one-policy">Steering one policy</h2> <ul> <li><strong>Text-conditioned behavior:</strong> classifier-free guidance strengthens the learned motion condition.</li> <li><strong>Test-time objectives:</strong> classifier guidance steers internal predicted states, enabling joystick control and other motion objectives.</li> <li><strong>Hardware execution:</strong> rolling denoising and computation-preserving runtime optimizations support a 20 ms control period on the onboard Jetson Orin NX.</li> </ul> <p>The paper reports a measured complete callback latency of <strong>16.790 ms median and 19.383 ms p95</strong> on the Jetson Orin NX. Only generated actions are applied to the robot; predicted states remain inside the policy for guidance.</p> <p> · · </p>