Robot vision method improves precise reaching with wrist camera input
Robot-Aligned 3D Gaussian Splatting for Precise Reaching with Wrist-Camera VLA Policies
Robotics
Summary
Robots often struggle to accurately reach for small parts in cluttered electronic waste because it’s hard to create perfect digital models of these complex items. The authors developed a new approach that uses a special 3D imaging technique called Gaussian Splatting to rebuild the workspace from the robot’s wrist camera view. This lets the robot learn to reach precisely by training on many realistic images and smooth motion paths without needing extra cameras. Their method helps robots better position their arms before touching objects, improving success in tasks like disassembling motherboards.
What this means in practice
- •For robotics engineers: Train robots to reach precisely in cluttered environments using only wrist camera views and synthetic data from 3D Gaussian Splatting reconstructions.
- •For electronic recycling operators: Use improved robotic reaching for automated disassembly of densely packed electronic components to increase efficiency and reduce manual labor.
- •For industrial automation integrators: Deploy wrist-camera-based vision systems for precise manipulation tasks without reliance on external cameras or expensive digital twins.$Commercial implications: Enables creation of commercial robotic systems for precise assembly or disassembly using only onboard cameras, reducing hardware costs.
Authors
Zuoxu Wang, Xiao Liang
Abstract
Efficient robotic data collection remains a major bottleneck for vision-language-action (VLA) policy learning, especially for tasks requiring accurate end-effector positioning and camera-action alignment. This challenge is particularly severe in electronic-waste disassembly, where robots operate in cluttered assemblies of densely packed, visually similar components. Full digital-twin simulation is difficult because many parts are thin, reflective, or mechanically complex, making accurate object-level meshes and contact models difficult to obtain. In this work, we focus on pre-contact reaching and propose a 3D Gaussian Splatting (3DGS)-based data generation pipeline for wrist-camera VLA adaptation. Our method reconstructs the real workspace with 3DGS and renders diverse wrist-camera observations from robot-aligned poses. These observations are paired with smooth, structured reaching trajectories to provide robot-aligned visual-action supervision for VLA fine-tuning. We further augment the VLA policy with an instruction-conditioned reach verifier that estimates target proximity and provides reliable termination before physical interaction. Experiments in motherboard disassembly demonstrate precise VLA-based reaching using wrist-camera observations alone, without an external side-view camera. With $π$0.5, 1,500 generated trajectories achieve 80.0\% precise-reaching success across three targets, compared with 58.7\% using 150 teleoperated demonstrations, while requiring less than 10 minutes of active human capture. Project website: https://001-wang.github.io/PrecisionReach-GS/