Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2609.30959

VisTacAlign: Co-Training Dexterous Policies on Tactile Human and Robot Demonstrations

Julien Poffet·Matthew Strong·Ankush Dhawan·Baiyu Shi·Shalika Neelaveni·Yujia Yuan·Zhenan Bao·Monroe Kennedy III
View original source ↗Read source document ↗Export metadata ↓

Abstract

Human demonstrations are a cheap source of data for dexterous manipulation, but co-training a robot policy on them requires closing the human--robot gap in every modality the policy consumes. We present VisTacAlign, a framework for co-training 3D-visual-tactile dexterous policies on human and robot demonstrations. Glove-tracked human hand motion is retargeted to a 17-DoF tactile robot hand with a one-time fingertip correction. The human hand is then erased from both stereo views and replaced by a posed robot-hand mesh painted with pixels from robot recordings, and a real-time stereo foundation model is re-run on the composite, so the human point clouds carry the same stereo errors and visibility as the robot ones. Finally, a capacitive tactile glove is aligned to the robot's fingertip sensors in its signal space, giving one interpretable per-finger force representation. A diffusion transformer consumes point-cloud, proprioceptive, and per-finger tactile tokens. On three real-world tasks requiring precise force -- Lego assembly, plucking strawberries of varying size, and activating and lifting a power drill -- adding aligned human demonstrations to existing robot data improves over robot-only policies, and ablations show that both tactile input and visual alignment are necessary. Project page: https://vis-tac-align.github.io

Explore related subjects

cs.RO

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Julien Poffet, Matthew Strong, Ankush Dhawan, Baiyu Shi, Shalika Neelaveni, Yujia Yuan, Zhenan Bao, Monroe Kennedy III. 2026-09-25. VisTacAlign: Co-Training Dexterous Policies on Tactile Human and Robot Demonstrations. https://arxiv.org/abs/2609.30959

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Structured-Diffuser: Diffusion with Task-Conditioned Structured Priors for Motion Planning

We propose Structured-Diffuser, a diffusion planner that embeds task and motion structure directly into the noise model. Unlike standard diffusion-based planners that rely on zero-mean, isotropic Gaussian corruption, we introduce task-conditioned structured Gaussians whose means and covariances are derived from Gaussian Process Motion Planning (GPMP), explicitly encoding trajectory smoothness and task semantics in the prior. We first formulate diffusion under a task-conditioned, non-isotropic Gaussian prior with closed-form forward and posterior expressions. Building on this formulation, our hierarchical design separates prior instantiation from trajectory denoising. At the upper level, sparse task-centric key states and timings are obtained, which instantiate a structured Gaussian prior (mean and covariance). At the lower level, the full trajectory is denoised under this prior, treating the upper-level outputs as noisy observations. Experiments across three motion-planning tasks show improved task success and training efficiency, with additional gains in data efficiency, trajectory smoothness, and position--velocity consistency where evaluated. Ablation studies further show that explicitly structuring the corruption process provides benefits beyond neurally conditioning the denoising network alone. Overall, our approach concentrates the prior's probability mass around task-relevant, temporally structured trajectories. We additionally demonstrate deployment on a physical G1 humanoid.

cs.RO↗
arXiv2026

SANDO: Safe Autonomous Trajectory Planning for Dynamic Unknown Environments

This paper presents SANDO, a safe trajectory planner for 3D dynamic unknown environments. Existing soft-constraint planners are fast but do not guarantee collision-free paths, while hard-constraint methods typically ensure safety at the cost of longer computation. SANDO addresses this trade-off through three contributions. First, a heat map-based A* global planner steers the path away from high-risk regions, and a spatiotemporal safe flight corridor (STSFC) generator produces time-layered polytopes that inflate obstacles only by their worst-case reachable set at each time layer, rather than over the entire horizon. Second, trajectory optimization is formulated as a mixed-integer quadratic program with hard collision-avoidance constraints, and variable elimination reduces the number of decision variables. Third, a formal safety analysis establishes collision-free guarantees under explicit velocity-bound, size-bound, and estimation-error assumptions. Ablation studies confirm that variable elimination yields up to 7.4 times faster optimization and that STSFCs are critical for feasibility in dense dynamic environments. In simulations against state-of-the-art methods, SANDO achieves a 100% success rate across all forest and dynamic benchmark difficulty levels with no constraint violations, and perception-only experiments demonstrate the full perception-to-planning pipeline. Hardware experiments with fully onboard planning, perception, and localization demonstrate six safe flights in static environments and twelve among dynamic obstacles.

cs.RO↗
arXiv2026

Cloak: Zero-Shot Cross-Embodiment Manipulation by Masking the End-Effector from the VLA

We present Cloak, a training recipe that endows a Vision-Language-Action (VLA) model with zero-shot cross-embodiment transfer by cloaking the end-effector from its own wrist camera. The end-effector occupies a large and consistent region of the wrist view and masking it allows for embodiment-agnostic visual reasoning. Cloak renders a mask in simulation from the robot's known geometry, accurately and in real time, with no segmentation or generative models. During training, we augment the mask so the model generalizes to embodiments unseen at training time. We demonstrate the recipe with Cloak-VLA, a VLA trained with Cloak on a single parallel-jaw gripper dataset. No data of new embodiments is ever collected. Cloak-VLA transfers zero-shot to various unseen embodiments, including another gripper, another arm, and a five-fingered hand, while preserving the source embodiment's performance. By decoupling the wrist view from its own embodiment, Cloak allows data to outlive the hardware it was collected on.

cs.RO↗