Variational Neural Belief Parameterizations for Robust Dexterous Grasping under Multimodal Uncertainty

2026-04-28

AI Summary

This work addresses the stochastic failures in dexterous grasping caused by contact uncertainty, perceptual noise, and external disturbances by proposing a risk-sensitive variational inference framework. The approach models the posterior belief over implicit contact parameters and object poses using a differentiable Gaussian mixture model, enabling end-to-end gradient-based optimization through Gumbel-Softmax and location-scale reparameterization. To explicitly enhance robustness in worst-case scenarios, a Conditional Value-at-Risk (CVaR) surrogate objective is incorporated. In simulation, the method significantly improves grasp success rates while reducing planning time by nearly an order of magnitude. Real-world experiments demonstrate faster, higher-quality tactile interactions and a risk calibration error below 0.14, substantially outperforming existing baselines.

Abstract

Contact variability, sensing uncertainty, and external disturbances make grasp execution stochastic. Expected-quality objectives ignore tail outcomes and often select grasps that fail under adverse contact realizations. Risk-sensitive POMDPs address this failure mode, but many use particle-filter beliefs that scale poorly, obstruct gradient-based optimization, and estimate Conditional Value-at-Risk (CVaR) with high-variance approximations. We instead formulate grasp acquisition as variational inference over latent contact parameters and object pose, representing the belief with a differentiable Gaussian mixture. We use Gumbel-Softmax component selection and location-scale reparameterization to express samples as smooth functions of the belief parameters, enabling pathwise gradients through a differentiable CVaR surrogate for direct optimization of tail robustness. In simulation, our variational neural belief improves robust grasp success under contact-parameter uncertainty and exogenous force perturbations while reducing planning time by roughly an order of magnitude relative to particle-filter model-predictive control. On a serial-chain robot arm with a multifingered hand, we validate grasp-and-lift success under object-pose uncertainty against a Gaussian baseline. Both methods succeed on the tested perturbations, but our controller terminates in fewer steps and less wall-clock time while achieving a higher tactile grasp-quality proxy. Our learned belief also calibrates risk more accurately, keeping mean absolute calibration error below 0.14 across tested simulation regimes, compared with 0.58 for a Cross-Entropy Method planner.

Problem

• robust grasping
• multimodal uncertainty
• risk-sensitive planning
• tail risk
• dexterous manipulation

Innovation

• variational inference
• differentiable Gaussian mixture
• risk-sensitive grasping
• Conditional Value-at-Risk (CVaR)
• pathwise gradients