Do Embodied Agents Dream of Pixelated Sheep: Embodied Decision Making using Language Guided World Modelling

Kolby Nottingham, Prithviraj Ammanabrolu, Alane Suhr, Yejin Choi, Hannaneh Hajishirzi, Sameer Singh, Roy Fox
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:26311-26325, 2023.

Abstract

Reinforcement learning (RL) agents typically learn tabula rasa, without prior knowledge of the world. However, if initialized with knowledge of high-level subgoals and transitions between subgoals, RL agents could utilize this Abstract World Model (AWM) for planning and exploration. We propose using few-shot large language models (LLMs) to hypothesize an AWM, that will be verified through world experience, to improve sample efficiency of RL agents. Our DECKARD agent applies LLM-guided exploration to item crafting in Minecraft in two phases: (1) the Dream phase where the agent uses an LLM to decompose a task into a sequence of subgoals, the hypothesized AWM; and (2) the Wake phase where the agent learns a modular policy for each subgoal and verifies or corrects the hypothesized AWM. Our method of hypothesizing an AWM with LLMs and then verifying the AWM based on agent experience not only increases sample efficiency over contemporary methods by an order of magnitude but is also robust to and corrects errors in the LLM, successfully blending noisy internet-scale information from LLMs with knowledge grounded in environment dynamics.

Cite this Paper


BibTeX
@InProceedings{pmlr-v202-nottingham23a, title = {Do Embodied Agents Dream of Pixelated Sheep: Embodied Decision Making using Language Guided World Modelling}, author = {Nottingham, Kolby and Ammanabrolu, Prithviraj and Suhr, Alane and Choi, Yejin and Hajishirzi, Hannaneh and Singh, Sameer and Fox, Roy}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {26311--26325}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/nottingham23a/nottingham23a.pdf}, url = {https://proceedings.mlr.press/v202/nottingham23a.html}, abstract = {Reinforcement learning (RL) agents typically learn tabula rasa, without prior knowledge of the world. However, if initialized with knowledge of high-level subgoals and transitions between subgoals, RL agents could utilize this Abstract World Model (AWM) for planning and exploration. We propose using few-shot large language models (LLMs) to hypothesize an AWM, that will be verified through world experience, to improve sample efficiency of RL agents. Our DECKARD agent applies LLM-guided exploration to item crafting in Minecraft in two phases: (1) the Dream phase where the agent uses an LLM to decompose a task into a sequence of subgoals, the hypothesized AWM; and (2) the Wake phase where the agent learns a modular policy for each subgoal and verifies or corrects the hypothesized AWM. Our method of hypothesizing an AWM with LLMs and then verifying the AWM based on agent experience not only increases sample efficiency over contemporary methods by an order of magnitude but is also robust to and corrects errors in the LLM, successfully blending noisy internet-scale information from LLMs with knowledge grounded in environment dynamics.} }
Endnote
%0 Conference Paper %T Do Embodied Agents Dream of Pixelated Sheep: Embodied Decision Making using Language Guided World Modelling %A Kolby Nottingham %A Prithviraj Ammanabrolu %A Alane Suhr %A Yejin Choi %A Hannaneh Hajishirzi %A Sameer Singh %A Roy Fox %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-nottingham23a %I PMLR %P 26311--26325 %U https://proceedings.mlr.press/v202/nottingham23a.html %V 202 %X Reinforcement learning (RL) agents typically learn tabula rasa, without prior knowledge of the world. However, if initialized with knowledge of high-level subgoals and transitions between subgoals, RL agents could utilize this Abstract World Model (AWM) for planning and exploration. We propose using few-shot large language models (LLMs) to hypothesize an AWM, that will be verified through world experience, to improve sample efficiency of RL agents. Our DECKARD agent applies LLM-guided exploration to item crafting in Minecraft in two phases: (1) the Dream phase where the agent uses an LLM to decompose a task into a sequence of subgoals, the hypothesized AWM; and (2) the Wake phase where the agent learns a modular policy for each subgoal and verifies or corrects the hypothesized AWM. Our method of hypothesizing an AWM with LLMs and then verifying the AWM based on agent experience not only increases sample efficiency over contemporary methods by an order of magnitude but is also robust to and corrects errors in the LLM, successfully blending noisy internet-scale information from LLMs with knowledge grounded in environment dynamics.
APA
Nottingham, K., Ammanabrolu, P., Suhr, A., Choi, Y., Hajishirzi, H., Singh, S. & Fox, R.. (2023). Do Embodied Agents Dream of Pixelated Sheep: Embodied Decision Making using Language Guided World Modelling. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:26311-26325 Available from https://proceedings.mlr.press/v202/nottingham23a.html.

Related Material