Publications

Collision-Aware Target-Driven Object Grasping in Constrained Environments

Published in IEEE International Conference on Robotics and Automation (ICRA), 2021

We propose a novel Collision-Aware Reachability Predictor (CARP) for 6-DoF grasping systems that learns to estimate the collision-free probabilities from partial/noisy observations and significantly improves grasping efficiency in challenging environments.

Recommended citation: X Lou, Y Yang, C Choi, Collision-Aware Target-Driven Object Grasping in Constrained Environments, IEEE International Conference on Robotics and Automation (ICRA) 2021. https://arxiv.org/pdf/2104.00776.pdf

Learning Visual Affordances with Target-Orientated Deep Q-Network to Grasp Objects by Harnessing Environmental Fixtures

Published in IEEE International Conference on Robotics and Automation (ICRA), 2021

Real-world objects may not be graspable with a single parallel gripper, but only with harnessing environment fixtures (e.g., walls, furniture, heavy objects). A slide-to-wall action is proposed as well as the Target-Oriented Deep Q-Network (TO-DQN) to efficiently learn visual affordance maps (i.e., Q-maps) to guide robot actions.

Recommended citation: Hengyue Liang, Xibai Lou, Yang Yang and Changhyun Choi, Learning Visual Affordances with Target-Orientated Deep Q-Network to Grasp Objects by Harnessing Environmental Fixtures, ICRA 2021. https://arxiv.org/pdf/1910.03781.pdf

Attribute-Based Robotic Grasping with One-Grasp Adaptation

Published in IEEE International Conference on Robotics and Automation (ICRA), 2021

We introduce an end-to-end learning method of attribute-based robotic grasping with one-grasp adaptation capability. Our approach fuses the embeddings of a workspace image and a query text using a gated-attention mechanism and learns to predict instance grasping affordances.

Recommended citation: Yang Yang, Yuanhao Liu, Hengyue Liang, Xibai Lou and Changhyun Choi, Attribute-Based Robotic Grasping with One-Grasp Adaptation, ICRA 2021. https://arxiv.org/pdf/2104.02271.pdf

Learning to generate 6-dof grasp poses with reachability awareness

Published in IEEE International Conference on Robotics and Automation (ICRA), 2020

A voxel-based deep 3D Convolutional Neural Network (3D CNN) that generates feasible 6-DoF grasp poses in unrestricted workspace with reachability awareness. The reachability predictor evaluates if the grasp pose is reachable by exploiting large-scale synthetic datasets.

Recommended citation: X Lou, Y Yang, C Choi, Learning to generate 6-dof grasp poses with reachability awareness, IEEE International Conference on Robotics and Automation (ICRA) 2020, 1532-1538 . https://arxiv.org/pdf/1910.06404.pdf