Motion Memory: Leveraging Past Experiences to Accelerate Future Motion Planning

TL;DR

Motion Memory addresses the inefficiency of solving motion-planning problems from scratch by leveraging a robot's past planning experiences through an experience augmentation pipeline and representation learning. It hallucinates planning problems to augment dataset $\mathcal{D}^*$ and learns a latent space via a triplet loss that clusters environments by the optimality of past plans; a new problem is encoded as $l_{N+1}=e_\theta(S_{N+1})$ and matched to centroid $c_i$ to select $i^*=\arg\min_i \|l_{N+1}-c_i\|$. The retrieved plan $P_{i^*}$ acts as a seed or guide for planning in both closed-box and open-box configurations. Experiments across three planners and three problem classes show speedups up to $89\%$ with increasing deployment experience, demonstrating both generality and potential for continual improvement in real-world robotics.

Abstract

When facing a new motion-planning problem, most motion planners solve it from scratch, e.g., via sampling and exploration or starting optimization from a straight-line path. However, most motion planners have to experience a variety of planning problems throughout their lifetimes, which are yet to be leveraged for future planning. In this paper, we present a simple but efficient method called Motion Memory, which allows different motion planners to accelerate future planning using past experiences. Treating existing motion planners as either a closed or open box, we present a variety of ways that Motion Memory can contribute to reduce the planning time when facing a new planning problem. We provide extensive experiment results with three different motion planners on three classes of planning problems with over 30,000 problem instances and show that planning speed can be significantly reduced by up to 89% with the proposed Motion Memory technique and with increasing past planning experiences.

Figures (6)

• Figure 1: Traditional motion planners require significant amount of effort to plan from scratch (left), such as large amount of samples or iterations (illustrated in green); Motion Memory utilizes past planning experiences to accelerate future planning when facing new planing problems (right).
• Figure 2: Motion Memory: Using Past Planning Experiences, Environment Generation, and Representation Learning to Accelerate Future Planning.
• Figure 3: Different Planning Problem Classes : Curves (Left), Random (Middle), Trap (Right)
• Figure 4: Different Motion Memory configurations improve three planners in three classes of motion planning problems.
• Figure 5: Improvement with Increasing Experiences.