Peg-in-hole assembly in unknown environments is a challenging task due to onboard sensor errors, which result in uncertainty and variations in task parameters such as the hole position and orientation. Meta Reinforcement Learning (Meta RL) has been proposed to mitigate this problem as it learns how to quickly adapt to new tasks with different parameters. However, previous approaches either depend on a sample-inefficient procedure or human demonstrations to perform the task in the real world. Our work modifies the data used by the Meta RL agent and uses simple features that can be easily measured in the real world even with an uncalibrated camera. We further adapt the Meta RL agent to use data from a force/torque sensor, instead of the camera, to perform the assembly, using a small amount of training data. Finally, we propose a fine-tuning method that consistently and safely adapts to out-of-distribution tasks with parameters that differ by a factor of 10 from the training tasks. Our results demonstrate that the proposed data modification significantly enhances the training and adaptation efficiency and enables the agent to achieve 100% success in tasks with different hole positions and orientations. Experiments on a real robot confirm that both camera- and force/torque sensor-equipped agents achieve 100% success in tasks with unknown hole positions, matching their simulation performance and validating the approach's robustness and applicability. Compared to the previous work with sample-inefficient adaptation, our proposed methods are 10 times more sample-efficient in the real-world tasks.

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.

Deep reinforcement learning could be used to learn dexterous robotic policies but it is challenging to transfer them to new robots with vastly different hardware properties. It is also prohibitively expensive to learn a new policy from scratch for each robot hardware due to the high sample complexity of modern state-of-the-art algorithms. We propose a novel approach called Hardware Conditioned Policies where we train a universal policy conditioned on a vector representation of robot hardware. We considered robots in simulation with varied dynamics, kinematic structure, kinematic lengths and degrees-of-freedom. First, we use the kinematic structure directly as the hardware encoding and show great zero-shot transfer to completely novel robots not seen during training. For robots with lower zero-shot success rate, we also demonstrate that fine-tuning the policy network is significantly more sample-efficient than training a model from scratch. In tasks where knowing the agent dynamics is important for success, we learn an embedding for robot hardware and show that policies conditioned on the encoding of hardware tend to generalize and transfer well. Videos of experiments are available at: https://sites.google.com/view/robot-transfer-hcp.