Search papers, labs, and topics across Lattice.
The paper introduces Self-Correcting VLA (SC-VLA), a novel vision-language-action model that improves physical grounding by integrating sparse world imagination with online action refinement. SC-VLA uses auxiliary predictive heads to forecast task progress and trajectory trends, guiding the policy to encode short-term physical evolution. By reshaping progress-dependent dense rewards based on predicted sparse future states, SC-VLA achieves state-of-the-art performance on robot manipulation tasks, demonstrating higher task throughput and success rates compared to existing methods in both simulation and real-world settings.
By intrinsically guiding action refinement through sparse imagination, SC-VLA achieves SOTA performance in robot manipulation tasks, outperforming existing methods by a significant margin in both simulation and real-world settings.
Standard vision-language-action (VLA) models rely on fitting statistical data priors, limiting their robust understanding of underlying physical dynamics. Reinforcement learning enhances physical grounding through exploration yet typically relies on external reward signals that remain isolated from the agent's internal states. World action models have emerged as a promising paradigm that integrates imagination and control to enable predictive planning. However, they rely on implicit context modeling, lacking explicit mechanisms for self-improvement. To solve these problems, we propose Self-Correcting VLA (SC-VLA), which achieve self-improvement by intrinsically guiding action refinement through sparse imagination. We first design sparse world imagination by integrating auxiliary predictive heads to forecast current task progress and future trajectory trends, thereby constraining the policy to encode short-term physical evolution. Then we introduce the online action refinement module to reshape progress-dependent dense rewards, adjusting trajectory orientation based on the predicted sparse future states. Evaluations on challenging robot manipulation tasks from simulation benchmarks and real-world settings demonstrate that SC-VLA achieve state-of-the-art performance, yielding the highest task throughput with 16% fewer steps and a 9% higher success rate than the best-performing baselines, alongside a 14% gain in real-world experiments. Code is available at https://github.com/Kisaragi0/SC-VLA.
