A variational Gaussian-mixture belief with Gumbel-Softmax and reparameterized sampling enables direct gradient optimization of tail-risk grasping objectives, improving success rates and cutting planning time versus particle-filter baselines.
Distri- butional Reinforcement Learning with Quantile Regression
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Variational Neural Belief Parameterizations for Robust Dexterous Grasping under Multimodal Uncertainty
A variational Gaussian-mixture belief with Gumbel-Softmax and reparameterized sampling enables direct gradient optimization of tail-risk grasping objectives, improving success rates and cutting planning time versus particle-filter baselines.