Real-time manipulability optimization with whole-arm obstacle avoidance for redundant manipulators via projection neural dynamics
Online motion of a redundant manipulator in a cluttered workspace has to reconcile end-effector tracking with two further goals: keeping every link clear of obstacles and staying away from kinematic singularities. Existing reactive and task-priority schemes usually address these goals separately or sacrifice tracking once whole-arm avoidance is added; the non-convex manipulability measure further burdens joint-space optimization. This article proposes the obstacle avoidance manipulability optimization (OAMO) scheme. A first-order Taylor expansion of manipulability with Tikhonov damping removes the gradient instability near singularities; tracking, manipulability enhancement, and link-level avoidance are then assembled into a single quadratic program (QP) in joint-velocity space with joint-velocity/joint-acceleration box constraints and distance-activated half-space constraints; the QP is solved by a projection neural dynamics recursion with an analytical proof of global convergence to the Karush–Kuhn–Tucker (KKT) optimum. On a 6-degree of freedom redundant manipulator (UR5e in our simulation), three case studies are reported: fixed-position reconfiguration, closed-contour tracking, and an obstacle-laden spiral. They show that OAMO consistently raises manipulability, preserves link-to-obstacle clearance, achieves a 1 0 − 3 m -level mean tracking error, and runs at a millisecond-scale cost per step, suitable for online motion control.
Authors
- Ranulfo BEZERRA (ORCID: https://orcid.org/0000-0001-6017-5724)
- Kimitoshi Yamazaki
- Tong Wu
Publication Details
- Journal
- Frontiers in Robotics and AI
- Published
- 2026-11-16
- DOI
- https://doi.org/10.3389/frobt.2026.1915338
- Primary Topic
- Robotic Mechanisms and Dynamics
- Type
- article
- Field-Weighted Citation Impact
- 0.00