Task-Driven Manipulation with Reconfigurable Parallel Robots

TL;DR

This work addresses manipulation with a reconfigurable parallel robot, ReachBot, in challenging environments by jointly optimizing stance and tension. The stance planner (a mixed-integer convex program) selects boom placements to maximize the task wrench space about a nominal pose in $SE(3)$, while the tension planner (a convex program) allocates boom tensions to realize the desired wrench $w \in \mathbb{R}^6$ under probabilistic grasping. Key contributions include a task-polytope based robustness objective, a pose-uncertainty handling heuristic, a multi-pose extension, and fast online tension computation, all validated by Monte-Carlo simulations showing enlarged manipulation workspace and improved robustness compared to a naïve baseline. The method enables reliable manipulation tasks such as sample extraction with in-situ catalyst and tool use in caves or lava tubes, and generalizes to other cable-driven morphologies, advancing autonomous scientific missions on the Moon, Mars, and beyond.

Abstract

ReachBot, a proposed robotic platform, employs extendable booms as limbs for mobility in challenging environments, such as martian caves. When attached to the environment, ReachBot acts as a parallel robot, with reconfiguration driven by the ability to detach and re-place the booms. This ability enables manipulation-focused scientific objectives: for instance, through operating tools, or handling and transporting samples. To achieve these capabilities, we develop a two-part solution, optimizing for robustness against task uncertainty and stochastic failure modes. First, we present a mixed-integer stance planner to determine the positioning of ReachBot's booms to maximize the task wrench space about the nominal point(s). Second, we present a convex tension planner to determine boom tensions for the desired task wrenches, accounting for the probabilistic nature of microspine grasping. We demonstrate improvements in key robustness metrics from the field of dexterous manipulation, and show a large increase in the volume of the manipulation workspace. Finally, we employ Monte-Carlo simulation to validate the robustness of these methods, demonstrating good performance across a range of randomized tasks and environments, and generalization to cable-driven morphologies. We make our code available at our project webpage, https://stanfordasl.github.io/reachbot_manipulation/

Figures (7)

• Figure 1: ReachBot performs a sample-extraction manipulation task in a cave environment. By optimally planning (i) where to place the booms and (ii) the tensions in each boom, we ensure that this task is completed even with uncertain pose and wrench estimation, and stochastic failure modes.
• Figure 2: Concept of operations. (A): Upon approaching an area of scientific interest, ReachBot's perception system identifies where the task needs to be performed, estimates a task wrench, identifies candidate grasp sites in the area, and estimates the quality of each site. (B): The stance planner considers the grasp sites and determines the optimal placement of each boom, for robust task execution. (C): After each boom is attached, the perception system can identify the quality of each grasp and determine the expected pull-force distribution for each site. (D): Using this grasp quality information, the tension planner determines the tensions on each boom to safely achieve the target task wrench.
• Figure 3: Relationship between the ReachBot stance, wrench space, and task polytope. The X/Y force space, a subset of the wrench space for a 2D ReachBot, is plotted, along with the task polytope and the task ellipsoid. (A): With no specified task, this ReachBot stance can resist arbitrary disturbance forces about the origin of the force space, with magnitudes up to the radius of the ball. (B): With a specified task ellipsoid or polytope, this stance can successfully execute the task even with some uncertainty, since the center of the ellipsoid/polytope is contained in the hull. (C): The task-ellipsoid/polytope metric can also indicate that a task is feasible, even when ReachBot is in a stance which is notably not in force closure (i.e. the origin of the space is not contained in the interior of the hull).
• Figure 4: Impact of the task-based stance optimization on the ReachBot wrench space. Here, we show the optimal ReachBot stances and corresponding wrench space (decomposed into the spaces of pure forces and torques) for two different tasks. By changing the task definition from a ball (left) to an ellipsoid biased towards forces along the z direction (right), the stance planner adjusts the wrench space to maximize the size of the inscribed task ellipsoid. This enables ReachBot to be robust against uncertainty in the wrench or set of wrenches experienced during manipulation.
• Figure 5: Comparison of two possible ReachBot stances for a set of possible sites. Left: a naïve configuration, simply directing the booms outwards from the body. This configuration has good control over the applied force, but is very weak in applying torque, since each boom has no effective lever arm on the body. Right: an optimal configuration, with good control over both applied force and torque.