analiza aiqbench
This paper proposes a generative vision-language-action policy that conditions action generation on a trajectory vocabulary built from expert demonstrations, then adds a residual correction module to limit drift from action chunking. The reported gains are in success rate and stability under disturbances, with low-latency inference claimed. Readers should check how the results hold up on real hardware and across tasks beyond the paper's benchmarks.
MobiGuide: Vocabulary-Guided Online Recovery for Mobile Manipulation
arXiv:2604.01567v2 Announce Type: replace Abstract: A central challenge in mobile manipulation is preserving multiple plausible action modes while remaining reactive during execution. For example, grasping a bottle in a cluttered scene can be achieved through multiple physically feasible base-arm motion plans. Deterministic VLA policies trained with $L_1$ regression are efficient, but they tend to average across valid modes and may produce physically invalid actions. Diffusion-based generative policies can better represent multimodal action distributions, yet their proposals may be weakly constrained when generated from pure noise, and iterative sampling can increase inference cost. Action chunking helps amortize inference, but it also introduces partially open-loop execution, allowing small mismatches to accumulate into drift. We present MobiGuide, a generative VLA model for efficient mobile manipulation, where action generation is guided by a structured trajectory vocabulary. Constructed from expert demonstrations, this vocabulary encodes plausible whole-body motion patterns and provides structured priors for feasible action generation. Given the current observation and instruction, MobiGuide locally refines these reference trajectories into context-compatible action proposals and selects the most suitable trajectory for execution. This preserves multimodal action candidates while keeping generation constrained around feasible expert-like motion modes. To mitigate chunk-induced drift, we further introduce a lightweight residual correction module that performs high-frequency, per-step adjustments during rollout. Across diverse mobile manipulation tasks, MobiGuide improves success and stability under execution disturbances while maintaining low-latency inference.
Abstract:A central challenge in mobile manipulation is preserving multiple plausible action modes while remaining reactive during execution. For example, grasping a bottle in a cluttered scene can be achieved through multiple physically feasible base-arm motion plans. Deterministic VLA policies trained with $L_1$ regression are efficient, but they tend to average across valid modes and may produce physically invalid actions. Diffusion-based generative policies can better represent multimodal action distributions, yet their proposals may be weakly constrained when generated from pure noise, and iterative sampling can increase inference cost. Action chunking helps amortize inference, but it also introduces partially open-loop execution, allowing small mismatches to accumulate into drift. We present MobiGuide, a generative VLA model for efficient mobile manipulation, where action generation is guided by a structured trajectory vocabulary. Constructed from expert demonstrations, this vocabulary encodes plausible whole-body motion patterns and provides structured priors for feasible action generation. Given the current observation and instruction, MobiGuide locally refines these reference trajectories into context-compatible action proposals and selects the most suitable trajectory for execution. This preserves multimodal action candidates while keeping generation constrained around feasible expert-like motion modes. To mitigate chunk-induced drift, we further introduce a lightweight residual correction module that performs high-frequency, per-step adjustments during rollout. Across diverse mobile manipulation tasks, MobiGuide improves success and stability under execution disturbances while maintaining low-latency inference.
Robotics (cs.RO)
arXiv:2604.01567 [cs.RO]
(or arXiv:2604.01567v2 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2604.01567 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 311 cuvinte
Subiecte conexe
Praxis: Distilling Physical Interaction Priors from Egocentric Videos for Generalizable Whole-Body Manipulation
arXiv · 28 sept. 2026
Digital States 2026: AI Moves From Experiment to Infrastructure
govtech.com · 24 sept. 2026
Hierarchical Residual Policy Learning for Real-World Mobile Manipulation with Sparse Human Guidance
arXiv · 15 oct. 2026
LiveLaw Supreme Court Half-Yearly Digest: January - June 2026 [Part 2]
livelaw.in · 13 sept. 2026