Vision-Language-Action (VLA) models frequently encounter challenges in generalizing to real-world environments due to inherent discrepancies between observation and action spaces. Although training data are collected from diverse camera perspectives, the models typically predict end-effector poses within the robot base coordinate frame, resulting in spatial inconsistencies. To mitigate this limitation, we introduce the Observation-Centric VLA (OC-VLA) framework, which grounds action predictions directly in the camera observation space. Leveraging the camera’s extrinsic calibration matrix, OC-VLA transforms end-effector poses from the robot base coordinate system into the camera coordinate system, thereby unifying prediction targets across heterogeneous viewpoints. This lightweight, plug-and-play strategy ensures robust alignment between perception and action, substantially improving model resilience to camera viewpoint variations. The proposed approach is readily compatible with existing VLA architectures, requiring no substantial modifications. Comprehensive evaluations on both simulated and real-world robotic manipulation tasks demonstrate that OC-VLA accelerates convergence, enhances task success rates, and improves cross-view generalization.

While recent vision-language-action models trained on diverse robot datasets exhibit promising generalization capabilities with limited in-domain data, their reliance on compact action heads to predict discretized or continuous actions constrains adaptability to heterogeneous action spaces. We present Dita, a scalable framework that leverages Transformer architectures to directly denoise continuous action sequences through a unified multimodal diffusion process. Departing from prior methods that condition denoising on fused embeddings via shallow networks, Dita employs in-context conditioning -- enabling fine-grained alignment between denoised actions and raw visual tokens from historical observations. This design explicitly models action deltas and environmental nuances. By scaling the diffusion action denoiser alongside the Transformer's scalability, Dita effectively integrates cross-embodiment datasets across diverse camera perspectives, observation scenes, tasks, and action spaces. Such synergy enhances robustness against various variances and facilitates the successful execution of long-horizon tasks. Evaluations across extensive benchmarks demonstrate state-of-the-art or comparative performance in simulation. Notably, Dita achieves robust real-world adaptation to environmental variances and complex long-horizon tasks through 10-shot finetuning, using only third-person camera inputs. The architecture establishes a versatile, lightweight and open-source baseline for generalist robot policy learning. Project Page: https://robodita.github.io.

Human-robot object handovers are essential for robots to effectively serve human needs in various domains of human–robot interaction and collaboration, yet remain a significant challenge. Remarkable progress has been made by parallel-jaw gripper robots in grasp generation and motion planning for handovers, while few studies address this issue using anthropomorphic hands, necessitating the ability to handle higher collision probabilities and lower approaching space under occluded situations. In this article, we present a reactive human-to-robot dexterous handover framework for anthropomorphic hands. The closed-loop framework employs an effective collision detection and grasp selection approach to ensure safe and smooth motion in unstructured environments. We implement a handover system using a UR5 robot arm and a Schunk SVH Hand based on the presented framework, which can react to human motion during the handover process, generalize to diverse objects with different 6-DoF poses, and execute suitable grasp configurations. The generalizability, reliability, and efficiency of our method are demonstrated through the handover of 30 novel objects, a system ablation study for submodule evaluation, and a user study assessment involving eight participants.

Grasping is one of the most fundamental skills for humans to interact with objects. However, it remains a challenging problem for anthropomorphic hands, due to the lack of object affordance understanding and high-dimensional grasp planning. In this work, we propose an anthropomorphic hand grasping framework to learn realistic and reasonable grasps in cluttered scenes, which tackles the problem in three items: 1) graspable point segmentation; 2) hand grasp generation and 3) grasp optimization. Specifically, our method generates high-quality hand grasps efficiently without complete object models by learning graspable points, associated grasp configurations from observed point cloud in a parallel manner and optimizing predicted grasps based on hand-object contacts. Simulation experiments show that our model generates physical plausible grasps for the anthropomorphic hand effectively with over 70% success rate. Real-world experiments demonstrate that the model trained in simulation performs satisfactorily in real-world scenarios for unseen objects.

This article investigates the challenge of enabling multifinger hands to perform human-like functional grasping for various intentions. However, accomplishing functional grasping in real robot hands present many challenges, including handling generalization ability for kinematically diverse robot hands, generating intention-conditioned grasps for a large variety of objects, and incomplete perception from a single-view camera. In this work, we first propose a six-step functional grasp synthesis algorithm based on fine-grained contact modeling. With the fine-grained contact-based optimization and learned dense shape correspondence, the algorithm is adaptable to various objects of the same category and a wide range of multifinger hands using few demonstrations. Second, over 10 k functional grasps are synthesized to train our neural network, named DexFG-Net, which generates intention-conditioned grasps based on reconstructed object. Extensive experiments in the simulation and physical grasps indicate that the grasp synthesis algorithm can produce human-like functional grasp with robust stability and functionality, and the DexFG-Net can generate plausible and human-like intention-conditioned grasping postures for anthropomorphic hands.