Meta learning improves pushing of unseen objects with fast online updates
MetaPusher: Meta Learning and Planning for Nonprehensile Manipulation of Unseen Objects with Rapid Online Adaption
Robotics
Summary
Robots find it hard to push objects they've never seen before because they don't know how heavy or slippery these objects are just by looking. The authors created a system called MetaPusher that helps robots quickly learn about an object's unique properties while pushing it, adjusting their plan as they go. This method combines fast learning from interaction and smart planning so the robot doesn’t have to practice on each object beforehand. The authors tested MetaPusher in simulations and real-life situations and found it made fewer mistakes and succeeded up to 20% more often than other methods.
What this means in practice
- •For robotics engineers: Enable robots to push and manipulate new objects effectively without prior training data by rapidly adapting to object dynamics during the task.
- •For autonomous system developers: Improve autonomous manipulation in logistics or manufacturing by planning online with updated models that reflect changing object behaviors.
Authors
Donghyung Lee, Seyedali Golestaneh, Jaskrit Singh, Zhuoyun Zhong, Athanasios Kapoutsis, Constantinos Chamzas
Abstract
Manipulating previously unseen objects remains challenging, as their dynamics depend on latent physical properties, such as friction and mass distribution, that cannot be inferred from perception alone. Prior experience across objects can provide an initial estimate of unseen object dynamics, but this estimate remains uncertain and can degrade further during sim-to-real transfer. Adapting the dynamics through interaction can progressively refine the estimation, however, updating the model may invalidate the planned trajectory. Successful and efficient manipulation therefore requires both rapid dynamics adaptation and a planning strategy that can incorporate this evolution. In this work, we introduce MetaPusher, a meta-learning and adaptive planning framework for nonprehensile manipulation of unseen objects without prior object-specific interactions. A meta-learned dynamics model rapidly adapts from interactions during task execution, while an adaptive kinodynamic planner updates long-horizon plans by reusing and refining its existing search tree. This coupling enables manipulation and adaptation without a separate data collection phase. We evaluate MetaPusher on unseen objects in simulation and in sim-to-real scenarios, comparing against fine-tuning and active learning methods, MPPI-based control, and a reinforcement learning policy. It achieves lower prediction error and improves task success rate by up to 20%.