Learning Generalizable Visual Representations via Interactive Gameplay
Luca Weihs, Aniruddha Kembhavi, Kiana Ehsani, Sarah M. Pratt, Winson Han, Alvaro Herrasti, Eric Kolve, Dustin Schwenk, Roozbeh Mottaghi, Ali Farhadi
Abstract
A growing body of research suggests that embodied gameplay, prevalent not just in human cultures but across a variety of animal species including turtles and ravens, is critical in developing the neural flexibility for creative problem solving, decision making, and socialization. Comparatively little is known regarding the impact of embodied gameplay upon artificial agents. While recent work has produced agents proficient in abstract games, these environments are far removed from the real world and thus these agents can provide little insight into the advantages of embodied play. Hiding games, such as hide-and-seek, played universally, provide a rich ground for studying the impact of embodied gameplay on representation learning in the context of perspective taking, secret keeping, and false belief understanding. Here we are the first to show that embodied adversarial reinforcement learning agents playing Cache, a variant of hide-and-seek, in a high fidelity, interactive, environment, learn generalizable representations of their observations encoding information such as object permanence, free space, and containment. Moving closer to biologically motivated learning strategies, our agents' representations, enhanced by intentionality and memory, are developed through interaction and play. These results serve as a model for studying how facets of vision develop through interaction, provide an experimental framework for assessing what is learned by artificial agents, and demonstrates the value of moving from large, static, datasets towards experiential, interactive, representation learning.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
13 top-tier papers have cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- Selective Visual Representations Improve Convergence and Generalization for Embodied AI cites Auxiliary Tasks and Exploration Enable ObjectGoal Navigation
- Interpretation of Emergent Communication in Heterogeneous Collaborative Embodied Agents cites GridToPix: Training Embodied Agents with Minimal Supervision
- GridToPix: Training Embodied Agents with Minimal Supervision cites Interpretation of Emergent Communication in Heterogeneous Collaborative Embodied Agents
- Habitat 3.0: A Co-Habitat for Humans, Avatars, and Robots cites Emergent Tool Use From Multi-Agent Autocurricula
- Habitat 3.0: A Co-Habitat for Humans, Avatars, and Robots cites Interpretation of Emergent Communication in Heterogeneous Collaborative Embodied Agents
- Habitat 3.0: A Co-Habitat for Humans, Avatars, and Robots cites GridToPix: Training Embodied Agents with Minimal Supervision
- Learning About Objects by Learning to Interact with Them cites Momentum Contrast for Unsupervised Visual Representation Learning
- GridToPix: Training Embodied Agents with Minimal Supervision cites Emergent Tool Use From Multi-Agent Autocurricula
Cited by top-tier papers13
- Habitat 3.0: A Co-Habitat for Humans, Avatars, and RobotsXavier Puig, Eric Undersander, Andrew Szot, Mikael Dallaire Cote et al.ICLR 2024 · 252 citations
- Auxiliary Tasks and Exploration Enable ObjectGoal NavigationJoel Ye, Dhruv Batra, Abhishek Das, Erik WijmansICCV 2021 · 137 citations
- Interpretation of Emergent Communication in Heterogeneous Collaborative Embodied AgentsShivansh Patel, Saim Wani, Unnat Jain, Alexander G. Schwing et al.ICCV 2021 · 30 citations
- Selective Visual Representations Improve Convergence and Generalization for Embodied AIAinaz Eftekhar, Kuo-Hao Zeng, Jiafei Duan, Ali Farhadi et al.ICLR 2024 · 28 citations
- GridToPix: Training Embodied Agents with Minimal SupervisionUnnat Jain, Iou-Jen Liu, Svetlana Lazebnik, Aniruddha Kembhavi et al.ICCV 2021 · 25 citations
Builds on4
- Emergent Tool Use From Multi-Agent AutocurriculaBowen Baker, Ingmar Kanitscheider, Todor M. Markov, Yi Wu et al.ICLR 2020 · 751 citations
- Learning About Objects by Learning to Interact with ThemMartin Lohmann, Jordi Salvador, Aniruddha Kembhavi, Roozbeh MottaghiNeurIPS 2020 · 19 citations
- Dynamics-Aware EmbeddingsWilliam F. Whitney, Rajat Agarwal, Kyunghyun Cho, Abhinav GuptaICLR 2020
- Momentum Contrast for Unsupervised Visual Representation LearningKaiming He, Haoqi Fan, Yuxin Wu, Saining Xie et al.CVPR 2020
Related papers
- On the Critical Role of Conventions in Adaptive Human-AI CollaborationAndy Shih, Arjun Sawhney, Jovana Kondic, Stefano Ermon et al.ICLR 2021 · 46 citations
- Continuous Scene Representations for Embodied AISamir Yitzhak Gadre, Kiana Ehsani, Shuran Song, Roozbeh MottaghiCVPR 2022 · 40 citations
- Efficient RL via Disentangled Environment and Agent RepresentationsKevin Gmelin, Shikhar Bahl, Russell Mendonca, Deepak PathakICML 2023 · 14 citations
- Learning Multi-Object Positional Relationships via Emergent CommunicationYicheng Feng, Boshi An, Zongqing LuAAAI 2024 · 4 citations
- Simple Embodied Language Learning as a Byproduct of Meta-Reinforcement LearningEvan Zheran Liu, Sahaana Suri, Tong Mu, Allan Zhou et al.ICML 2023 · 4 citations
