Transferable Trial-Minimizing Progressive Peg-in-hole Model

Peg-in-hole is a fine-level manipulation task that requires highly precise location and the normal direction of the hole, which is beyond state-of-the-art object detectors’ capability. Therefore, we propose a novel method that trains a robot arm to progressively search for the right inserting pose through trials with both force feedback and visual inputs. Under a reinforcement learning (RL) framework, an agent trying to minimize the number of trials is learned based on the sequentially estimated relative poses with regard to the correct inserting pose. Moreover, our learned dynamics model is transferable. Thanks to our context-independent force and visual feature design, our pre-trained model can be finetuned efficiently for another unseen peg-in-hole case. Extensive experiments show the effectiveness of the proposed framework.

Authors

• Junfeng Ding
• Chen Wang 0033
• Cewu Lu

Keywords

• Visualization
• Force feedback
• Force
• Dynamics
• Reinforcement learning
• Manipulators
• Intelligent robots
• Context modeling
• Dynamic Model
• Object Detection
• Visual Features
• Visual Input
• Manipulation Tasks
• Robotic Arm
• Sequential Estimation
• Relative Pose
• Recurrent Neural Network
• Commons Attribution
• Multilayer Perceptron
• Transfer Model
• Optical Flow
• Pose Estimation
• Force Sensor
• Robot Manipulator
• Circular Hole
• Flow Map
• Extra Training
• Reinforcement Learning Agent
• Form Of A Heatmap
• Torque Force
• Successful Insertion
• Sensor Feedback
• Current Pose
• Edge Of The Hole
• Gradient Update
• Zero-shot
• Model-based Reinforcement Learning