Online Adaptation of Sampling-Based Motion Planning with Inaccurate Models

TL;DR

This work proposes a sampling-based motion planning approach that uses an estimate of the model error and online observations to correct the planning strategy at each new replanning and introduces the notion of context-awareness, which stores local environment information for each executed transition and avoids new transitions with context similar to previous unreliable ones.

Abstract

Robotic manipulation relies on analytical or learned models to simulate the system dynamics. These models are often inaccurate and based on offline information, so that the robot planner is unable to cope with mismatches between the expected and the actual behavior of the system (e.g., the presence of an unexpected obstacle). In these situations, the robot should use information gathered online to correct its planning strategy and adapt to the actual system response. We propose a sampling-based motion planning approach that uses an estimate of the model error and online observations to correct the planning strategy at each new replanning. Our approach adapts the cost function and the sampling bias of a kinodynamic motion planner when the outcome of the executed transitions is different from the expected one (e.g., when the robot unexpectedly collides with an obstacle) so that future trajectories will avoid unreliable motions. To infer the properties of a new transition, we introduce the notion of context-awareness, i.e., we store local environment information for each executed transition and avoid new transitions with context similar to previous unreliable ones. This is helpful for leveraging online information even if the simulated transitions are far (in the state-and-action space) from the executed ones. Simulation and experimental results show that the proposed approach increases the success rate in execution and reduces the number of replannings needed to reach the goal.

Figures (4)

• Figure 1: A 7-degree-of-freedom manipulator carrying a weight with uncertain tracking control.
• Figure 2: Sketch of the proposed method. From left to right: (i) the robot plans and executes a trajectory minimizing the expected error. It stops when it deviates too much from the nominal path. (ii) every time it stops, we observe the actual error and the context of the executed transitions. (iii) we update the cost function and the sampling bias to avoid transitions similar to the unreliable ones. (iV) the new solution will likely avoid such transitions.
• Figure 3: Scenarios for the numerical analysis in Sec. \ref{['sec:2d-example']}.
• Figure 4: Examples of executions of two trajectories planned with CTX-RRT (top) and MAB-RRT (bottom).