PolarNet: 3D Point Clouds for Language-Guided Robotic Manipulation

Abstract

The ability for robots to comprehend and execute manipulation tasks based on natural language instructions is a long-term goal in robotics. The dominant approaches for language-guided manipulation use 2D image representations, which face difficulties in combining multi-view cameras and inferring precise 3D positions and relationships. To address these limitations, we propose a 3D point cloud based policy called PolarNet for language-guided manipulation. It leverages carefully designed point cloud inputs, efficient point cloud encoders, and multimodal transformers to learn 3D point cloud representations and integrate them with language instructions for action prediction. PolarNet is shown to be effective and data efficient in a variety of experiments conducted on the RLBench benchmark. It outperforms state-of-the-art 2D and 3D approaches in both single-task and multi-task learning. It also achieves promising results on a real robot.

Accepted to CoRL 2023. Project website: https://www.di.ens.fr/willow/research/polarnet/

References in corpus (12)

• Open3D: A Modern Library for 3D Data Processing
• PaLM-E: An Embodied Multimodal Language Model
• VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection
• CLIPort: What and Where Pathways for Robotic Manipulation
• BC-Z: Zero-Shot Task Generalization with Robotic Imitation Learning
• Perceiver-Actor: A Multi-Task Transformer for Robotic Manipulation
• Instruct2Act: Mapping Multi-modality Instructions to Robotic Actions with Large Language Model
• Interactive Language: Talking to Robots in Real Time
• Language Conditioned Spatial Relation Reasoning for 3D Object Grounding
• DexPoint: Generalizable Point Cloud Reinforcement Learning for Sim-to-Real Dexterous Manipulation
• Frame Mining: a Free Lunch for Learning Robotic Manipulation from 3D Point Clouds
• ToolFlowNet: Robotic Manipulation with Tools via Predicting Tool Flow from Point Clouds