ROBBRIEF
The arXiv preprint merges equality and inequality constraints into one geometry. The paper reports 100% on its own test suite, with hard collision checks still separate.
Robot motion planning, the way a robotic arm decides where to move next, usually runs two separate systems. One forces the arm to follow exact motion rules: a joint trajectory, a tool path, a hand orientation. The other tests each candidate move against obstacles, throwing most moves away. The collision tests are where the time goes.
A new arXiv preprint, RMRRT: Riemannian Barrier Metric RRT for Inequality-Aware Steering on Equality Manifolds, merges those two systems into one local geometry. Inequality constraints become a soft "barrier" field that pushes the planner away from obstacles. Equality constraints stay exact via projection into the tangent space. The new planner steers and samples neighbors inside that single combined metric, so it wastes less effort deciding how to dodge.
The paper reports a 100% success rate on its constrained-manipulation test suite, in simulation and in a real-world arm setup, and shorter planning times than the constrained-planning baselines it picked. Ablations show fewer rejected samples and shorter paths with the proposed metric.
The qualifier: 100% is on the authors' chosen tasks and baselines, not all robotic manipulation. Hard collision avoidance is still enforced by a separate validity check; the new metric biases exploration, it does not replace the safety filter. The submission is anonymous (project page at rmrrt-anonymous.github.io, code at anonymous.4open.science) and is an arXiv preprint, not peer-reviewed.
For contact-rich assembly, in-hand manipulation, or lab/surgical-adjacent robotics, where both reliability and planning speed matter, the question is whether an implementer picks it up. Watch that, not the headline number.