Mixture of Horizons in Action Chunking

TL;DR

This paper introduces a mixture of horizons (MoH) strategy for vision-language-action models in robotic manipulation, combining different action chunk lengths to improve both local control and long-term foresight, leading to better performance and efficiency.

Contribution

The paper proposes a novel MoH approach that processes multiple action horizons in parallel, enabling dynamic horizon selection and joint exploitation of short-term and long-term information.

Findings

MoH improves performance on complex tasks in simulation and real-world.

MoH achieves 2.5x higher throughput with maintained accuracy.

State-of-the-art success rate of 99% on LIBERO with fewer training iterations.

Abstract

Vision-language-action (VLA) models have shown remarkable capabilities in robotic manipulation, but their performance is sensitive to the $\textbf{action chunk length}$ used during training, termed $\textbf{horizon}$. Our empirical study reveals an inherent trade-off: longer horizons provide stronger global foresight but degrade fine-grained accuracy, while shorter ones sharpen local control yet struggle on long-term tasks, implying fixed choice of single horizons being suboptimal. To mitigate the trade-off, we propose a $\textbf{mixture of horizons (MoH)}$ strategy. MoH rearranges the action chunk into several segments with different horizons, processes them in parallel with a shared action transformer, and fuses outputs with a light linear gate. It has three appealing benefits. 1) MoH exploits long-term foresight and short-term precision jointly within a single model, improving both performance and generalizability to complex tasks. 2) MoH is plug-and-play for full-attention action modules with minimal training or inference overhead. 3) MoH enables dynamic inference with adaptive horizons, which selects stable actions through cross-horizon consensus, achieving 2.5$\times$ higher throughput than baselines while preserving superior performance. Extensive experiments over flow-based policies $\pi_0$, $\pi_{0.5}$, and one-step regression policy $\pi_{\text{reg}}$ demonstrate that MoH yields consistent and significant gains on both simulations and real-world tasks. Notably, under mixed-task setting, $\pi_{0.5}$ with MoH reaches a new state-of-the-art with 99$\%$ average success rate on LIBERO after only $30k$ training iterations. Project page: https://github.com/Timsty1/MixtureOfHorizons