Variational Neural Belief Parameterizations for Robust Dexterous Grasping under Multimodal Uncertainty
2026-04-28 • Robotics
RoboticsMachine Learning
AI summaryⓘ
The authors studied how robots can better grasp objects even when things like touch and position are uncertain. Instead of using traditional methods that can be slow and sometimes fail in tricky situations, they developed a new way to estimate what might happen using a smoother, faster model. This new method helps the robot plan grasps that are more reliable and quick in both computer simulations and real-world tests. It also measures risk more accurately, helping the robot avoid failure in harder cases. Overall, their approach makes robotic grasping more efficient and dependable under uncertainty.
grasp executionstochastic processesConditional Value-at-Risk (CVaR)variational inferenceGaussian mixture modelsGumbel-Softmaxrobotic manipulationmodel-predictive controlbelief representationtactile sensing
Authors
Clinton Enwerem, Shreya Kalyanaraman, John S. Baras, Calin Belta
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.