A Peg-in-hole Task Strategy for Holes in Concrete

TL;DR

The paper addresses robust peg-in-hole insertions into concrete holes characterized by high friction and varied surface finish. It introduces a data-driven approach where a DNN is trained via reinforcement learning to predict next search positions, while detaching the peg between searches to mitigate friction; the peg displacement input $D_Z$ is explicitly incorporated to improve generalization. Empirical results show up to 96.1% success with average times below 12.5 s, and strong generalization to random initial positions and different peg types, with displacement input boosting performance and saliency analyses confirming its importance. The findings suggest practical applicability for construction automation and potential extension to other cylindrical insertions in brittle, high-friction materials.

Abstract

A method that enables an industrial robot to accomplish the peg-in-hole task for holes in concrete is proposed. The proposed method involves slightly detaching the peg from the wall, when moving between search positions, to avoid the negative influence of the concrete's high friction coefficient. It uses a deep neural network (DNN), trained via reinforcement learning, to effectively find holes with variable shape and surface finish (due to the brittle nature of concrete) without analytical modeling or control parameter tuning. The method uses displacement of the peg toward the wall surface, in addition to force and torque, as one of the inputs of the DNN. Since the displacement increases as the peg gets closer to the hole (due to the chamfered shape of holes in concrete), it is a useful parameter for inputting in the DNN. The proposed method was evaluated by training the DNN on a hole 500 times and attempting to find 12 unknown holes. The results of the evaluation show the DNN enabled a robot to find the unknown holes with average success rate of 96.1% and average execution time of 12.5 seconds. Additional evaluations with random initial positions and a different type of peg demonstrate the trained DNN can generalize well to different conditions. Analyses of the influence of the peg displacement input showed the success rate of the DNN is increased by utilizing this parameter. These results validate the proposed method in terms of its effectiveness and applicability to the construction industry.

Figures (10)

• Figure 1: Proposed method. $P_{hole}$ is the rough hole position obtained by, for example, a vision-based detection algorithm; $F_x$, $F_y$, and $F_z$ are the forces on X, Y, Z axes; $M_x$ and $M_y$ are the moments on the X and Y axes; $D_Z$ is peg displacement on the Z axis; and $P_{next}$ is the next search position obtained by the DNN. Subscript $th$ refers to threshold.
• Figure 2: Usage example of proposed method. $P_{Z,init}$ is initial offset position from the wall, $D_Z$ is peg displacement from $P_{Z,init}$, and $D_{x,y}$ is peg displacement to the next search position in X (left/right) and Y (up/down).
• Figure 3: Deep Q-Learning architecture
• Figure 4: Experimental setup for inserting anchor bolt
• Figure 5: Holes for training (in blue) and evaluations (in orange)