GeoPredict: Leveraging Predictive Kinematics and 3D Gaussian Geometry for Precise VLA Manipulation
This addresses the need for more precise and reliable 3D-aware manipulation in robotics, representing an incremental advancement by enhancing existing VLA frameworks with predictive modules.
The paper tackles the problem of unreliable 3D reasoning in Vision-Language-Action models for robotic manipulation by proposing GeoPredict, which integrates predictive kinematic and geometric priors, resulting in consistent performance improvements over baselines on benchmarks like RoboCasa Human-50 and LIBERO.
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.