Visual Foresight With a Local Dynamics Model
This work addresses the data inefficiency problem for robotics researchers and practitioners in manipulation tasks, offering an incremental improvement by integrating model-based planning with model-free learning.
The paper tackles the problem of time-consuming training and large data requirements in model-free policy learning for long-horizon manipulation tasks by proposing a Local Dynamics Model (LDM) that efficiently learns state-transition functions for manipulation primitives. The result shows that LDM is more sample-efficient, outperforms other model architectures, and when combined with planning, surpasses other model-based and model-free policies on challenging manipulation tasks in simulation.
Model-free policy learning has been shown to be capable of learning manipulation policies which can solve long-time horizon tasks using single-step manipulation primitives. However, training these policies is a time-consuming process requiring large amounts of data. We propose the Local Dynamics Model (LDM) which efficiently learns the state-transition function for these manipulation primitives. By combining the LDM with model-free policy learning, we can learn policies which can solve complex manipulation tasks using one-step lookahead planning. We show that the LDM is both more sample-efficient and outperforms other model architectures. When combined with planning, we can outperform other model-based and model-free policies on several challenging manipulation tasks in simulation.