Existing research that addressed cable manipulation relied on two-fingered grippers, which make it difficult to perform similar cable manipulation tasks that humans perform. However, unlike dexterous manipulation of rigid objects, the development of dexterous cable manipulation skills in robotics remains underexplored due to the unique challenges posed by a cable's deformability and inherent uncertainty. In addition, using a dexterous hand introduces specific difficulties in tasks, such as cable grasping, pulling, and in-hand bending, for which no dedicated task definitions, benchmarks, or evaluation metrics exist. Furthermore, we observed that most existing dexterous hands are designed with structures identical to humans', typically featuring only one thumb, which often limits their effectiveness during dexterous cable manipulation. Lastly, existing non-task-specific methods did not have enough generalization ability to solve these cable manipulation tasks or are unsuitable due to the designed hardware. We have three contributions in real-world dexterous cable manipulation in the following steps: (1) We first defined and organized a set of dexterous cable manipulation tasks into a comprehensive taxonomy, covering most short-horizon action primitives and long-horizon tasks for one-handed cable manipulation. This taxonomy revealed that coordination between the thumb and the index finger is critical for cable manipulation, which decomposes long-horizon tasks into simpler primitives. (2) We designed a novel five-fingered hand with 25 degrees of freedom (DoF), featuring two symmetric thumb-index configurations and a rotatable joint on each fingertip, which enables dexterous cable manipulation. (3) We developed a demonstration collection pipeline for this non-anthropomorphic hand, which is difficult to operate by previous motion capture methods.

Reinforcement Learning (RL) based methods have been increasingly explored for robot learning. However, RL based methods often suffer from low sampling efficiency in the exploration phase, especially for long-horizon manipulation tasks, and generally neglect the semantic information from the task level, resulted in a delayed convergence or even tasks failure. To tackle these challenges, we propose a Temporal-Logic-guided Hybrid policy framework (HyTL) which leverages three-level decision layers to improve the agent's performance. Specifically, the task specifications are encoded via linear temporal logic (LTL) to improve performance and offer interpretability. And a waypoints planning module is designed with the feedback from the LTL-encoded task level as a high-level policy to improve the exploration efficiency. The middle-level policy selects which behavior primitives to execute, and the low-level policy specifies the corresponding parameters to interact with the environment. We evaluate HyTL on four challenging manipulation tasks, which demonstrate its effectiveness and interpretability. Our project is available at: https://sites.google.com/view/hytl-0257/.

Real-world robotic tasks stretch over extended horizons and encompass multiple stages. Learning long-horizon manipulation tasks, however, is a long-standing challenge, and demands decomposing the overarching task into several manageable subtasks to facilitate policy learning and generalization to unseen tasks. Prior task decomposition methods require task-specific knowledge, are computationally intensive, and cannot readily be applied to new tasks. To address these shortcomings, we propose Universal Visual Decomposer (UVD), an off-the-shelf task decomposition method for visual long horizon manipulation using pre-trained visual representations designed for robotic control. At a high level, UVD discovers subgoals by detecting phase shifts in the embedding space of the pre-trained representation. Operating purely on visual demonstrations without auxiliary information, UVD can effectively extract visual subgoals embedded in the videos, while incurring zero additional training cost on top of standard visuomotor policy training. Goal-conditioned policies learned with UVD-discovered subgoals exhibit significantly improved compositional generalization at test time to unseen tasks. Furthermore, UVD-discovered subgoals can be used to construct goal-based reward shaping that jump-starts temporally extended exploration for reinforcement learning. We extensively evaluate UVD on both simulation and real-world tasks, and in all cases, UVD substantially outperforms baselines across imitation and reinforcement learning settings on in-domain and out-of-domain task sequences alike, validating the clear advantage of automated visual task decomposition within the simple, compact UVD framework.