Reinforcement learning (RL) has been successfully applied to a wealth of robot manipulation tasks and continuous control problems. However, it is still limited to industrial applications and suffers from three major challenges: sample inefficiency, real data collection, and the gap between simulator and reality. In this paper, we focus on the practical application of RL for robot assembly in the real world. We apply enlightenment learning to improve the proximal policy optimization, an on-policy model-free actor-critic reinforcement learning algorithm, to train an agent in Cartesian space using the proprioceptive information. We introduce enlightenment learning incorporated via pretraining, which is beneficial to reduce the cost of policy training and improve the effectiveness of the policy. A human-like assembly trajectory is generated through a two-step method with segmenting objects by locations and iterative closest point for pretraining. We also design a sim-to-real controller to correct the error while transferring to reality. We set up the environment in the MuJoCo simulator and demonstrated the proposed method on the recently established The National Institute of Standards and Technology (NIST) gear assembly benchmark. The paper introduces a unique framework that enables a robot to learn assembly tasks efficiently using limited real-world samples by leveraging simulations and visual demonstrations. The comparative experiment results indicate that our approach surpasses other baseline methods in terms of training speed, success rate, and efficiency.

Robotic peg-hole insertion operations can play a very important role in manufacturing industry because it is a common requirement in the manufacturing process and, it requires high precision and high speed.

In general, this operation can be divided into two processes: Search process to engage the peg and the hole and, insertion process.

The search process is critical in the assembly operation. It can be defined as a process where the angular.and the translational errors between the peg and the hole are reduced until insertion can occur.

In this paper the authors present a mathematical model of geometric and kinematic behaviour of an original passive compliant device provided with two rotation centres, called DCR–LAI system. This device is designed for a robotic assembly of parts with very small tolerances including a chamfer at the hole. The given modélisation may be used as a decision aid for the choice of a compliant device with regard to characteristics of parts being assembled.

In this paper a pneumatic vibratory wrist operated with a PWM controller is developed for robotic assembly. In the vibratory assembly system, the vibratory wrist can perform random search motion of a hole to compensate the position error at the early stage of an insertion process. Since the vibration characteristics of the wrist, such as the amplitude and trajectory of the vibration, are critical to assembly performance, they are experimentally investigated for various system controller parameters. In addition, a series of insertion experiments are performed to evaluate the assembly performance of the proposed wrist. The results show that within a wide range of operating conditions the wrist vibration can effectively compensate for large positioning errors when this wrist is used for a chamferless peg-in-hole task.

In this paper we develop a compilant system that permits robotic assembly of chamferless pieces. The idea is to absorb the positioning error between parts to be inserted by giving one of them a planar random movement. An actuator consisting of two axes (X and y) operated by an electromagnetic System is fitted to the work table; when its inputs are pseudo-random binary signais (P.R.B.S.) random motion is obtained. The trajectories of the actuator are analysed depending upon the P.R.B.S. parameters and a peg-in-a-hole assembly task is carried out. Experimental results show that large positioning errors can be compensated for chamferless insertions.

This paper describes a compact assembly sequencer for automatic robotic assembly using a configuration space-based path planner. Connection graph of the assembly is constrained by the feasibility information made available from the analysis of local disassembly paths tried by the robot's path planner. Free space of the robot performing assembly is explicitly used to generate the disassembly constraints. We show with the help of a ball pen assembly how the sequencing is performed with this representation.

This paper focuses on developing a planning strategy for robotic assembly. At first, channel and junction are defined by using half-spaces, and free-space inside the female part is approximately decomposed by channels and junctions. Then, a simple and efficient algorithm to find the assembly path of the male part is developed, in which any path between the shortest and safest paths can be easily found by just changing the clearance gap between the male and female parts. Next, the robot arm is considered in the path planning, in which a feasible grasp angle region is obtained to avoid a collision between the robot arm and the female part during the assembly process. An optimum grasping angle can be found in the feasible grasping angle region by applying a proper performance index. Finally, a simple robotic assembly using the algorithm is numerically demonstrated.

The complexity and uncertainty of the cross-sectional shape of the parts to be mated is one of the main reasons that misalignment between them occurs in assembly processes. Misalignment cannot only give rise to assembly failure but also cause damage to the parts or a robot due to large contact force. Therefore, misalignment sensing and compensation is essential for successful assembly operation. In this paper, we propose a novel misalignment estimation and compensation method which does not need any advance information on the cross-sectional shapes of the mating parts. This method utilizes a $\varphi-r$ transformation and an M-estimation pattern matching technique with misalignment images of a peg and a hole taken by an omni-directional visual sensing system during assembly. At every sampling instant during assembly action, it furnishes information on the relative position and orientation between the mating parts, and thus helps to estimate and compensate any possible misalignment between them. Also, a series of experiments are performed with a couple of peg-in-hole tasks, and the results are discussed. The experimental results show that the proposed method is effective for misalignment compensation in robotic assembly even though there is no prior information on part geometry and the images are very noisy.

It is an established fact that a majority of a product's cost is determined by its design. Hence, effort should be directed to achieve a lower cost design without sacrificing the original functionality. A computer-aided product redesign system is proposed to provide help in generating assembly-oriented redesign by taking the robotic assembly constraints into consideration. The system operates in two phases: In the first phase, an objective evaluation of the input design is performed to determine whether there is a need for redesign. During the second phase, the focus is to aid the designer in searching for feasible design alternatives. Three quantitative measures have been proposed to evaluate the conformity of an input design to a list of DFA design guidelines. If the input design received a bad evaluation results on any of the three measures, a search for feasible redesign suggestions is initiated to derive suitable redesign suggestions.