# Action Chunking

Vision-language-action models usually predict a whole future action chunk in one pass, and adjacent chunks lack explicit temporal consistency constraints, so actions can jump at chunk boundaries. ODEAction recasts generation as continuous evolution from the current action: it learns an action transition field and integrates it with an ordinary differential equation to produce subsequent actions, instead of jointly predicting a fixed-length chunk. Average success rate rises from 96.85% to 98.25% on LIBERO and from 15.28% to 24.72% on EBench, a 61.8% relative gain; on five RoboTwin tasks it reaches 84.4% average success against the 82.4% baseline.

Vision-language-action models usually predict a whole future action chunk in one pass, and adjacent chunks lack explicit temporal consistency constraints, so actions can jump at chunk boundaries. ODEAction recasts generation as continuous evolution from the current action: it learns an action transition field and integrates it with an ordinary differential equation to produce subsequent actions, instead of jointly predicting a fixed-length chunk. Average success rate rises from 96.85% to 98.25% on LIBERO and from 15.28% to 24.72% on EBench, a 61.8% relative gain; on five RoboTwin tasks it reaches 84.4% average success against the 82.4% baseline.

In humanoid manipulation, a large vision-language-action backbone is slow, so synchronous execution acts on stale observations and control frequency falls as the backbone grows. DASH splits the policy into a slow vision-language model and a fast flow-matching action head running as three concurrent processes over LCM, so the control rate is set by the action head alone. On a RoboTwin benchmark that charges inference latency in simulated time, DASH holds 20 Hz from 0.8B to 8B backbones while the synchronous rate falls from 9.7 Hz to 4.1 Hz, with higher success at every size; on a real dexterous humanoid, pick-and-place succeeds in 23 of 24 trials versus 16 of 24 for the synchronous pipeline.

In humanoid manipulation, a large vision-language-action backbone is slow, so synchronous execution acts on stale observations and control frequency falls as the backbone grows. DASH splits the policy into a slow vision-language model and a fast flow-matching action head running as three concurrent processes over LCM, so the control rate is set by the action head alone. On a RoboTwin benchmark that charges inference latency in simulated time, DASH holds 20 Hz from 0.8B to 8B backbones while the synchronous rate falls from 9.7 Hz to 4.1 Hz, with higher success at every size; on a real dexterous humanoid, pick-and-place succeeds in 23 of 24 trials versus 16 of 24 for the synchronous pipeline.

Online reinforcement-learning adaptation of long-chunk robot policies previously optimized high-dimensional action sequences directly, which demanded many samples and made real-robot adaptation costly. ExRL instead adapts a scalar decision, how many steps of each predicted action chunk to execute, training a lightweight adapter with off-policy reinforcement learning while the base policy stays frozen. Across simulation benchmarks and four real manipulation tasks it improves diffusion policies, ACT, several vision-language-action models and a world action model with limited online interaction, approaching perfect success on real cube stacking after about ten minutes of physical interaction.

Online reinforcement-learning adaptation of long-chunk robot policies previously optimized high-dimensional action sequences directly, which demanded many samples and made real-robot adaptation costly. ExRL instead adapts a scalar decision, how many steps of each predicted action chunk to execute, training a lightweight adapter with off-policy reinforcement learning while the base policy stays frozen. Across simulation benchmarks and four real manipulation tasks it improves diffusion policies, ACT, several vision-language-action models and a world action model with limited online interaction, approaching perfect success on real cube stacking after about ten minutes of physical interaction.

Autonomous lunar rover docking has relied on modular pipelines that split perception, mapping and planning, so perception errors propagate into planning. Researchers trained a vision-language-action policy with a Qwen3.5-4B backbone on safety-aware demonstrations from a privileged expert in the LuPNT lunar simulator, mapping a single monocular camera and a bird's-eye-view map directly to waypoints. Under dust and optical disturbances unseen in training, the policy achieved higher docking success in five of six disturbance conditions, with the largest advantage under the strongest perturbations; adaptive-horizon execution held 90% docking success while cutting policy inferences per successful scenario from 17.16 with a fixed 2 m horizon to 8.36.

Autonomous lunar rover docking has relied on modular pipelines that split perception, mapping and planning, so perception errors propagate into planning. Researchers trained a vision-language-action policy with a Qwen3.5-4B backbone on safety-aware demonstrations from a privileged expert in the LuPNT lunar simulator, mapping a single monocular camera and a bird's-eye-view map directly to waypoints. Under dust and optical disturbances unseen in training, the policy achieved higher docking success in five of six disturbance conditions, with the largest advantage under the strongest perturbations; adaptive-horizon execution held 90% docking success while cutting policy inferences per successful scenario from 17.16 with a fixed 2 m horizon to 8.36.