Learning Diffusion Policies for Robotic Manipulation of Timber Joinery under Fabrication Uncertainty

TL;DR

The paper tackles the challenge of robotic timber joinery under fabrication uncertainty by applying sensory-motor diffusion policy learning trained via behavior cloning on pose and force/torque data. A two-phase experimental design evaluates both performance in a deterministic static setting and robustness to randomized mortise offsets, yielding a 100% success rate in the static case and up to 75% average success under perturbations. Ablation studies highlight the critical role of force/torque feedback and indicate that a substantial number of demonstrations (roughly 200–400) are needed for reliable recovery behaviors under uncertainty. The work provides actionable guidelines for data collection and parameter tuning and demonstrates the potential for diffusion policies to generalize to other contact-rich construction and manufacturing tasks, advancing robotic construction under uncertainty toward safer and more efficient practice.

Abstract

Construction uncertainties such as fabrication inaccuracies and material imperfections pose a significant challenge to contact-rich robotic manipulation by hindering precise and robust assembly. In this paper, we explore the performance and robustness of diffusion policy learning as a promising solution for contact-sensitive robotic assembly at construction scale, using timber mortise and tenon joints as a case study. A two-phase study is conducted: first, to evaluate policy performance and applicability; second, to assess robustness in handling fabrication uncertainties simulated as randomized perturbations to the mortise position. The best-performing policy achieved a total average success rate of 75% with perturbations up to 10 mm, including 100% success in unperturbed cases. The results demonstrate the potential of sensory-motor diffusion policies to generalize to a wide range of complex, contact-rich assembly tasks across construction and manufacturing, advancing robotic construction under uncertainty and contributing to safer, more efficient building practices.

Figures (11)

• Figure 1: Overview of our method. A CNN-based diffusion policy is trained conditional on end-effector pose and F/T data observations to predict sequences of robot actions.
• Figure 2: Multi-robot setup consisting of two 6-axis industrial robotic arms.
• Figure 3: Gripper for holding the tenon piece equipped with anti-collision and F/T sensors.
• Figure 4: VR-based teleoperation system for collecting human expert demonstrations.
• Figure 5: Data collection and policy rollout workflow.