This paper presents a novel method for online learning of grasp predictions for robotic bin picking. The approach uses a unique exploration strategy to improve adaptation to unseen environment settings. The method is formulated as a reinforcement learning problem, allowing for the adaptation of both grasp reward prediction and grasp poses. The study employs uncertainty estimation schemes based on Bayesian Uncertainty Quantification and Distributional Ensembles. The results show significant improvement in performance compared to conventional online learning methods.
Publication date: 22 Sep 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2309.12038
