DayDreamer: World Models for Physical Robot Learning

By Philipp Wu, Alejandro Escontrela, Danijar Hafner, Ken Goldberg, Pieter Abbeel. Conference on Robot Learning, 2022

533 citations recorded by the source corpus.

To solve tasks in complex environments, robots need to learn from experience. Deep reinforcement learning is a common approach to robot learning but requires a large amount of trial and error to learn, limiting its deployment in the physical world. As a consequence, many advances in robot learning rely on simulators. On the other hand, learning inside of simulators fails to capture the complexity of the real world, is prone to simulator inaccuracies, and the resulting behaviors do not adapt to changes in the world. The Dreamer algorithm has recently shown great promise for learning from small amounts of interaction by planning within a learned world model, outperforming pure reinforcement learning in video games. Learning a world model to predict the outcomes of potential actions enables planning in imagination, reducing the amount of trial and error needed in the real environment. However, it is unknown whether Dreamer can facilitate faster learning on physical robots. In this paper, we apply Dreamer to 4 robots to learn online and directly in the real world, without simulators. Dreamer trains a quadruped robot to roll off its back, stand up, and walk from scratch and without resets in only 1 hour. We then push the robot and find that Dreamer adapts within 10 minutes to withstand perturbations or quickly roll over and stand back up. On two different robotic arms, Dreamer learns to pick and place multiple objects directly from camera images and sparse rewards, approaching human performance. On a wheeled robot, Dreamer learns to navigate to a goal position purely from camera images, automatically resolving ambiguity about the robot orientation. Using the same hyperparameters across all experiments, we find that Dreamer is capable of online learning in the real world, establishing a strong baseline. We release our infrastructure for future applications of world models to robot learning.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/25bc06b508b2c63b9faf77881e528530b147b988

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Reasoning with Language Model is Planning with World Model (2023, unknown): “Besides, the world model is also shown to be helpful to physical robot learning [62].”
A Survey on Vision--Language--Action Models for Embodied AI (2024, unknown): “DayDreamer [73] applies this method to physical robots performing real-world tasks.”
Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (2024, methodology): “Meanwhile, researchers have explored few-shot adaptation and real-world RL, either model-free (48, 53) or model-based (54), to update policies using real-world rollouts to further generalize policies to novel situations without accurate simulation.”
GAIA-1: A Generative World Model for Autonomous Driving (2023, methodology): “World models have been successfully applied to control tasks in both simulation [9, 10, 11, 12, 13] and to real-world robotics tasks [14, 15].”
Driving Into the Future: Multiview Visual Forecasting and Planning with World Model for Autonomous Driving (2023, methodology): “Learning world models in either game [19–21, 43, 47] or lab environments [12, 14, 67] has been widely studied.”
Extreme Parkour with Legged Robots (2023, methodology): “This has motivated the use of learned controllers trained with RL that can adapt to changes in dynamics [18, 34, 9, 46, 20, 35, 40] and also leverage elevation maps [25, 38, 32, 28, 29] for perceptive walking.”
Zero-Shot Robotic Manipulation with Pretrained Image-Editing Diffusion Models (2023, methodology): “For instance, [21, 37, 63, 60, 53, 22] train actionconditioned dynamics models and run RL in the model.”
Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models (2023, background): “…of the solution. branches of outcomes are explored during search, is widely used in many planning algorithms (Swiechowski et al., 2021; LaValle, 1998) and reinforcement learning (RL) (Hafner et al., 2019; Du et al., 2023; Wu et al., 2023) algorithms for its good exploration-exploitation trade-off.”
V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning (2025, unknown): “…in a learned representation space (Watter et al., 2015; Agrawal et al., 2016; Ha and Schmidhuber, 2018; Hafner et al., 2019b; Nair et al., 2022; Wu et al., 2023b; Tomar et al., 2024; Hu et al., 2024; Lancaster et al., 2024), or utilizing more structured representation spaces such as keypoint…”