Constrained Stein Variational Gradient Descent for Robot Perception, Planning, and Identification

TL;DR

This paper introduces two novel frameworks that extend Stein variational gradient descent to handle constrained optimization problems in robotics, enabling the learning of feasible distributions for motion planning, pose estimation, and identification.

Contribution

The paper presents new frameworks for applying constrained Stein variational gradient descent, supporting various constraints and demonstrating their effectiveness in robotics applications.

Findings

Learned collision-free robot motion plans

Estimated object poses with placement constraints

Generated feasible joint angle distributions

Abstract

Many core problems in robotics can be framed as constrained optimization problems. Often on these problems, the robotic system has uncertainty, or it would be advantageous to identify multiple high quality feasible solutions. To enable this, we present two novel frameworks for applying principles of constrained optimization to the new variational inference algorithm Stein variational gradient descent. Our general framework supports multiple types of constrained optimizers and can handle arbitrary constraints. We demonstrate on a variety of problems that we are able to learn to approximate distributions without violating constraints. Specifically, we show that we can build distributions of: robot motion plans that exactly avoid collisions, robot arm joint angles on the SE(3) manifold with exact table placement constraints, and object poses from point clouds with table placement constraints.