InternVLA-A1.5 combines native vision-language understanding, training-only latent foresight, and action generation to improve compositional generalization and long-horizon robotic manipulation.

EBench diagnoses generalist mobile manipulation policies with 26 Isaac Sim tasks across capability and generalization dimensions.

EruDiff refactors world knowledge in text-to-image diffusion models to improve synthesis from implicit prompts across scientific and commonsense knowledge benchmarks.

A unified VLA framework that combines understanding, visual foresight, and action generation for robust robotic manipulation in dynamic and static scenarios.

GraspLDP injects grasp priors into latent diffusion policy learning to improve grasp precision and generalization in both simulation and real-world manipulation.

An active perception method for grasp detection by introducing the neural graspness field, which models the grasp distribution of a scene.

Generalizing 6-DoF grasp detection framework with domain prior knowledge of robotic grasping.

We present a global-to-local method to address hybrid domain gaps in RGB and depth data and insufficient multi-modal feature alignment.

We build a depth guided learning framework, where both the RGB and depth images are fed and their features are combined to generate grasp proposals.

Focus on the problem of feature learning in the presence of scale imbalance for 6-DoF grasp detection and propose a novel approach to especially address the difficulty in dealing with small-scale samples.

We exploit boundary as a significant guidance for context aggregation to promote the overall semantic understanding of an image.