Comparing Task Simplifications to Learn Closed-Loop Object Picking Using Deep Reinforcement Learning

Abstract

Enabling autonomous robots to interact in unstructured environments with dynamic objects requires manipulation capabilities that can deal with clutter, changes, and objects' variability. This paper presents a comparison of different reinforcement learning-based approaches for object picking with a robotic manipulator. We learn closed-loop policies mapping depth camera inputs to motion commands and compare different approaches to keep the problem tractable, including reward shaping, curriculum learning and using a policy pre-trained on a task with a reduced action set to warmstart the full problem. For efficient and more flexible data collection, we train in simulation and transfer the policies to a real robot. We show that using curriculum learning, policies learned with a sparse reward formulation can be trained at similar rates as with a shaped reward. These policies result in success rates comparable to the policy initialized on the simplified task. We could successfully transfer these policies to the real robot with only minor modifications of the depth image filtering. We found that using a heuristic to warm-start the training was useful to enforce desired behavior, while the policies trained from scratch using a curriculum learned better to cope with unseen scenarios where objects are removed.

Review of Deep Reinforcement Learning-Based Object Grasping: Techniques, Open Challenges, and Recommendations

“…Robust object manipulation is a key component in all robotic applications which require interaction with the environment. In [249], the authors addressed the issue of closed-loop learning policies regarding the combined task of reaching, grasping, and lifting objects. The policies have a mapped depth image in their system, which is collected by a wrist-mounted camera, to the motion of the end-effector, and the gripper opening and closing commands.…”

Object Detection Recognition and Robot Grasping Based on Machine Learning: A Survey

“…Gu et al [150] proposed a new deep reinforcement learning algorithm based on deep Q-functions nonstrategy training that can adapt to complex 3D operation tasks. Breyer et al [151] proposed an object grabbing algorithm based on reinforcement learning. In this paper, the image collected by a depth camera is mapped to the closed-loop control strategy of motion command, and several different methods are compared to ensure the rationality of the algorithm.…”

“…Robot [147,151,156,184] Computer vision [185][186][187][188][189][190] Game [191][192][193] Autonomous driving [185,186]…”

Cooperative Multi-Agent Interaction and Evaluation Framework Considering Competitive Networks with Dynamic Topology Changes

“…The robots learn a closed-loop policy that maps depth camera inputs to motion commands and compare different approaches to make the problem easier to deal with, including the reward formation, curriculum learning, and the use of pre-trained policies with reduced work to pre-start tasks. Training the robots with heuristics helped achieve the desired behavior [22]. Collaborative robots are widely used in hybrid assembly tasks involving intelligent manufacturing.…”