Local Path Optimization in The Latent Space Using Learned Distance Gradient

TL;DR

This work tackles the bottleneck of replanning in constrained motion planning when using latent-space representations by introducing a local latent-space path optimization that leverages a learned minimum-distance gradient to push latent waypoints away from obstacles. The approach combines CVAE-based latent encoding, a minimum-distance predictor, and a neural validity checker within an LCBiRRT framework, enabling fast, constraint-aware planning with reduced online replanning. Across three planning scenarios, the method achieves faster planning times and higher success rates than state-of-the-art baselines, with performance gains especially pronounced in complex tasks. The results suggest that local latent-space optimization has strong practical potential for real-time, high-dimensional constrained manipulation, with future work pointing toward global optimization and faster validity checks.

Abstract

Constrained motion planning is a common but challenging problem in robotic manipulation. In recent years, data-driven constrained motion planning algorithms have shown impressive planning speed and success rate. Among them, the latent motion method based on manifold approximation is the most efficient planning algorithm. Due to errors in manifold approximation and the difficulty in accurately identifying collision conflicts within the latent space, time-consuming path validity checks and path replanning are required. In this paper, we propose a method that trains a neural network to predict the minimum distance between the robot and obstacles using latent vectors as inputs. The learned distance gradient is then used to calculate the direction of movement in the latent space to move the robot away from obstacles. Based on this, a local path optimization algorithm in the latent space is proposed, and it is integrated with the path validity checking process to reduce the time of replanning. The proposed method is compared with state-of-the-art algorithms in multiple planning scenarios, demonstrating the fastest planning speed

Figures (6)

• Figure 1: Schematic of the proposed method, by performing local path optimization in the latent space, the robot gets out of the obstacle area.
• Figure 2: The workflow of the proposed method.
• Figure 3: Schematic of the local path optimization process in the latent space.
• Figure 4: Three experimental scenarios and the robot motion paths.
• Figure 5: The spheres used to envelope the robot.