Learning Probabilistic Multi-Modal Actor Models for Vision-Based Robotic Grasping

• DOI: 10.48550/arxiv.1904.07319
• Publication Date: 2019-01-01
• Authors: Mengyuan Yan; Adrian Li; Mrinal Kalakrishnan; Peter Pástor
• Keywords: FOS: Computer and information sciences; Computer Science - Robotics; Computer Science - Machine Learning; 0209 industrial biotechnology; 02 engineering and technology; Robotics (cs.RO); Machine Learning (cs.LG)

Many previous works approach vision-based robotic grasping by training a value network that evaluates grasp proposals. These approaches require an optimization process at run-time to infer the best action from network. As result, inference time grows exponentially as dimension of space increases. We propose alternative method, directly neural density model approximate conditional distribution successful poses input images. construct combines Gaussian mixture and normalizing flows, which is able represent multi-modal, complex probability distributions. demonstrate on both simulation real robot proposed actor achieves similar performance compared using Cross-Entropy Method (CEM) for inference, top-down with 4 dimensional space. Our reduces 3 times state-of-the-art CEM method. believe models will play important role when scaling up these higher spaces.