Action Chunking
A policy output method in robot learning in which a policy network's single forward pass predicts a sequence of actions for several future steps, known as a chunk, rather than only the action for a single timestep. The robot executes the sequence in order before replanning, reducing the computation and jitter of high-frequency replanning.
Timeline (last 90 days)
Action chunking refers to predicting and executing multiple consecutive action steps at once rather than re-inferring at every step; temporal action quantization can be viewed as dynamically adjusting the chunk length.
Existing VLA accelerators (such as action chunking, visual token pruning, and quantized compilation) generally report only average task success rates, masking the risk of cascading failures that can occur during closed-loop execution.
All Facts (2)
Action chunking refers to predicting and executing multiple consecutive action steps at once rather than re-inferring at every step; temporal action quantization can be viewed as dynamically adjusting the chunk length.
50%UnverifiedExisting VLA accelerators (such as action chunking, visual token pruning, and quantized compilation) generally report only average task success rates, masking the risk of cascading failures that can occur during closed-loop execution.
50%