Spatial Action Maps for Mobile Manipulation

Abstract

Typical end-to-end formulations for learning robotic navigation involve predicting a small set of steering command actions (e.g., step forward, turn left, turn right, etc.) from images of the current state (e.g., a bird's-eye view of a SLAM reconstruction). Instead, we show that it can be advantageous to learn with dense action representations defined in the same domain as the state. In this work, we present "spatial action maps," in which the set of possible actions is represented by a pixel map (aligned with the input image of the current state), where each pixel represents a local navigational endpoint at the corresponding scene location. Using ConvNets to infer spatial action maps from state images, action predictions are thereby spatially anchored on local visual features in the scene, enabling significantly faster learning of complex behaviors for mobile manipulation tasks with reinforcement learning. In our experiments, we task a robot with pushing objects to a goal location, and find that policies learned with spatial action maps achieve much better performance than traditional alternatives.

arXiv commentTo appear at Robotics: Science and Systems (RSS), 2020. Project page: https://spatial-action-maps.cs.princeton.edu

References in corpus (20)

• End to End Learning for Self-Driving Cars
• Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments
• From Perception to Decision: A Data-driven Approach to End-to-end Motion Planning for Autonomous Ground Robots
• AI2-THOR: An Interactive 3D Environment for Visual AI
• Building Generalizable Agents with a Realistic and Rich 3D Environment
• Target-driven Visual Navigation in Indoor Scenes using Deep Reinforcement Learning
• MINOS: Multimodal Indoor Simulator for Navigation in Complex Environments
• Intention-Net: Integrating Planning and Deep Learning for Goal-Directed Autonomous Navigation
• CHALET: Cornell House Agent Learning Environment
• Learning to Act by Predicting the Future
• Gibson Env: Real-World Perception for Embodied Agents
• Robotic Pick-and-Place of Novel Objects in Clutter with Multi-Affordance Grasping and Cross-Domain Image Matching
• Playing Doom with SLAM-Augmented Deep Reinforcement Learning
• FollowNet: Robot Navigation by Following Natural Language Directions with Deep Reinforcement Learning
• Learning Exploration Policies for Navigation
• Form2Fit: Learning Shape Priors for Generalizable Assembly from Disassembly
• Grasping in the Wild:Learning 6DoF Closed-Loop Grasping from Low-Cost Demonstrations
• Learning to Move with Affordance Maps
• Deep Reinforcement Learning to Acquire Navigation Skills for Wheel-Legged Robots in Complex Environments
• AutOTranS: an Autonomous Open World Transportation System

Cited by in corpus (20)

• FlingBot: The Unreasonable Effectiveness of Dynamic Manipulation for Cloth Unfolding
• ReLMoGen: Leveraging Motion Generation in Reinforcement Learning for Mobile Manipulation
• From Screens to Scenes: A Survey of Embodied AI in Healthcare
• N${}^2$M${}^2$: Learning Navigation for Arbitrary Mobile Manipulation Motions in Unseen and Dynamic Environments
• Spatial Intention Maps for Multi-Agent Mobile Manipulation
• Learning to Set Waypoints for Audio-Visual Navigation
• VAT-Mart: Learning Visual Action Trajectory Proposals for Manipulating 3D ARTiculated Objects
• Learning 3D Dynamic Scene Representations for Robot Manipulation
• QDP: Learning to Sequentially Optimise Quasi-Static and Dynamic Manipulation Primitives for Robotic Cloth Manipulation
• Learning to Rearrange Deformable Cables, Fabrics, and Bags with Goal-Conditioned Transporter Networks
• Learning Pneumatic Non-Prehensile Manipulation with a Mobile Blower
• DeRi-Bot: Learning to Collaboratively Manipulate Rigid Objects via Deformable Objects
• SSCNav: Confidence-Aware Semantic Scene Completion for Visual Semantic Navigation
• Disentangled Planning and Control in Vision Based Robotics via Reward Machines
• Equivariant $Q$ Learning in Spatial Action Spaces
• Affordance as general value function: A computational model
• Learning Visual Affordances with Target-Orientated Deep Q-Network to Grasp Objects by Harnessing Environmental Fixtures
• Policy learning in SE(3) action spaces
• Tool as Embodiment for Recursive Manipulation
• Visual Perspective Taking for Opponent Behavior Modeling