[1904.03177] Structured agents for physical construction

Physical construction---the ability to compose objects, subject to physical dynamics, to serve some function---is fundamental to human intelligence. We introduce a suite of challenging physical construction tasks inspired by how children play with blocks, such as matching a target configuration, stacking blocks to connect objects together, and creating shelter-like structures over target objects. We examine how a range of deep reinforcement learning agents fare on these challenges, and introduce several new approaches which provide superior performance. Our results show that agents which use structured representations (e.g., objects and scene graphs) and structured policies (e.g., object-centric actions) outperform those which use less structured representations, and generalize better beyond their training when asked to reason about larger scenes. Model-based agents which use Monte-Carlo Tree Search also outperform strictly model-free agents in our most challenging construction problem...

4 mentions: @DeepMindAI@NogaRot@AISC_TO@luckflow
Date: 2019/06/11 17:15

Referring Tweets

@DeepMindAI This work is led by Victor Bapst and Alvaro Sanchez-Gonalez (also with @CarlDoersch, @neuro_kim, @pushmeet, @PeterWBattaglia, and @jhamrick). Read more here: https://t.co/c0w9TD3eOj
@NogaRot So many great talks by female researchers today! This is Structured agents for physical construction given by Jessica Hamrick @jhamrick . I have to read their graph NN overview. See more at poster #36 later today! Full paper: https://t.co/Y3iBUrZQeT #ICML2019 https://t.co/Ci3ylcSUex
@AISC_TO Structured Agents for Physical Construction: this work uses graph neural network to model relative object relations and physical properties in performing structure construction tasks: https://t.co/X8ydC7gWDL Video: https://t.co/ZSwsGkibbm