Object-centric Forward Modeling for Model Predictive Control
This addresses the challenge of robust robot control in dynamic environments, though it appears incremental as it builds on existing forward modeling approaches.
The paper tackles the problem of planning action sequences for robots to achieve distant goals by learning an object-centric forward model from random interaction data, which improves sample efficiency and prediction accuracy over implicit or pixel-space models in both simulation and real-world experiments.
We present an approach to learn an object-centric forward model, and show that this allows us to plan for sequences of actions to achieve distant desired goals. We propose to model a scene as a collection of objects, each with an explicit spatial location and implicit visual feature, and learn to model the effects of actions using random interaction data. Our model allows capturing the robot-object and object-object interactions, and leads to more sample-efficient and accurate predictions. We show that this learned model can be leveraged to search for action sequences that lead to desired goal configurations, and that in conjunction with a learned correction module, this allows for robust closed loop execution. We present experiments both in simulation and the real world, and show that our approach improves over alternate implicit or pixel-space forward models. Please see our project page (https://judyye.github.io/ocmpc/) for result videos.