TacDiffusion: Force-domain Diffusion Policy for Precise Tactile Manipulation

TL;DR

TacDiffusion introduces a diffusion-based policy that generates 6D force-domain actions for tactile manipulation in high-precision insertion tasks, enabling zero-shot transfer from a single demonstration. By integrating the diffusion policy with impedance control and a dynamic system-based filter, the approach achieves high robustness and real-time feasibility, reporting a 95.7% average success on novel tasks. The work also dissects the trade-off between diffusion-model size (inference speed) and accuracy, and demonstrates a 9.15% performance improvement through frequency alignment. Overall, TacDiffusion offers a practical framework for transferable tactile manipulation that can adapt to unseen high-precision insertions with improved efficiency and stability.

Abstract

Assembly is a crucial skill for robots in both modern manufacturing and service robotics. However, mastering transferable insertion skills that can handle a variety of high-precision assembly tasks remains a significant challenge. This paper presents a novel framework that utilizes diffusion models to generate 6D wrench for high-precision tactile robotic insertion tasks. It learns from demonstrations performed on a single task and achieves a zero-shot transfer success rate of 95.7% across various novel high-precision tasks. Our method effectively inherits the self-adaptability demonstrated by our previous work. In this framework, we address the frequency misalignment between the diffusion policy and the real-time control loop with a dynamic system-based filter, significantly improving the task success rate by 9.15%. Furthermore, we provide a practical guideline regarding the trade-off between diffusion models' inference ability and speed.

Figures (10)

• Figure 1: Network architecture of the noise estimator.
• Figure 2: Experiment Setup. The object grasped by the robot in the left figure is the training object: (a) Cuboid: A 35 mm × 25 mm × 60 mm dimensional cuboid (0.1 mm clearance). The four objects on the right are applied to validate the transferability: (b) Key: A 37 mm long key; (c) Cyl-S: A 50 mm long cylinder with a diameter of 20 mm (0.02 mm clearance); (d) Cyl-L: A cylinder with a length of 50 mm and diameter of 30 mm (0.025 mm clearance); (e) Prism: A 50 mm long octagonal prism with a side length of 11 mm (0.05 mm clearance).
• Figure 3: An example view of observations in the dataset.
• Figure 4: Training loss and validation loss. Validation is conducted every 5 epochs throughout the training process.
• Figure 5: Denoising process with model $DF_{3}$. From the top down, the red curves indicate the change in the diffused actions during the denoising process. The black refers to the corresponding ground truth.