GeoPredict: Leveraging Predictive Kinematics and 3D Gaussian Geometry for Precise VLA Manipulation

TL;DR

GeoPredict is a geometry-aware VLA framework that enhances robotic manipulation by integrating predictive kinematic and 3D Gaussian geometry modules, improving 3D reasoning and spatial accuracy.

Contribution

It introduces a trajectory prediction and 3D geometry module that serve as training supervision, enabling precise 3D manipulation without complex inference overhead.

Findings

Outperforms strong VLA baselines on multiple datasets.

Improves accuracy in geometry-intensive manipulation tasks.

Uses training-time supervision with depth rendering, lightweight inference.

Abstract

Vision-Language-Action (VLA) models achieve strong generalization in robotic manipulation but remain largely reactive and 2D-centric, making them unreliable in tasks that require precise 3D reasoning. We propose GeoPredict, a geometry-aware VLA framework that augments a continuous-action policy with predictive kinematic and geometric priors. GeoPredict introduces a trajectory-level module that encodes motion history and predicts multi-step 3D keypoint trajectories of robot arms, and a predictive 3D Gaussian geometry module that forecasts workspace geometry with track-guided refinement along future keypoint trajectories. These predictive modules serve exclusively as training-time supervision through depth-based rendering, while inference requires only lightweight additional query tokens without invoking any 3D decoding. Experiments on RoboCasa Human-50, LIBERO, and real-world manipulation tasks show that GeoPredict consistently outperforms strong VLA baselines, especially in geometry-intensive and spatially demanding scenarios.