A framework to generalize long-horizon extrinsic manipulation tasks from a single demonstration by retargeting the contact requirements to diverse object and environment configurations.
AdaDemo, a framework that actively and continuously expands the demonstration dataset, can progressively improve the performance of multi-task visual policies in a data-efficient manner.
This work proposes a unified framework for efficient task-oriented dexterous hand pose synthesis without human data, by introducing a novel, fast, accurate, and differentiable approach to estimating the Grasp Wrench Space and a novel task-oriented energy for optimizing dexterous hand poses.
RoboMP2 introduces a Goal-Conditioned Multimodal Perceptor (GCMP) and a Retrieval-Augmented Multimodal Planner (RAMP) to enhance the perception and planning capabilities of embodied agents by leveraging multimodal large language models.
This paper presents a novel decomposition of the infinite-horizon toss juggling movement into a sequential short-horizon trajectory optimization problem, identifying the critical constraints necessary for this dexterous dynamic manipulation task with switching contacts. The authors demonstrate stable juggling of up to 17 balls on two anthropomorphic manipulators, reaching the theoretical limits of toss juggling.
PreAfford utilizes a novel relay training paradigm and point-level affordance representation to enhance adaptability across a broad range of environments and object types, including those previously unseen, while maintaining compatibility with easy-to-grasp objects.
This article presents an approach for modeling the uncertainty of contact dynamics in order to synthesize robust manipulation behavior through open-loop pushing plans. The key contributions are: 1) deriving a prediction of the variance of object configurations upon contact, 2) introducing a contact prior for sampling candidate robot trajectories, and 3) proposing a sampling-based trajectory optimization algorithm that constrains solutions to be robust based on the predicted variance.
A novel collision-free generative diffusion model approach, APEX, is proposed to efficiently generate diverse and seamless trajectories for ambidextrous dual-arm robotic manipulation tasks.
A quadratic programming-based control approach is presented that enables tracking control of dual-arm robotic manipulators performing planned simultaneous impacts, avoiding error peaking and input steps.
Large Language Models can be effectively employed to coordinate the control of bimanual robots in accomplishing complex long-horizon manipulation tasks by generating sequential or simultaneous control policies based on the current task state.