Complicated assembly processes can be described as a sequence of two main activities: grasping and insertion. While general solutions are common in industry, insertion is still only applicable to small subsets problems, mainly ones involving simple shapes fixed locations which the variations not taken into consideration. Recently, RL approaches with prior knowledge (e.g., LfD or residual policy) have been adopted. However, these might problematic contact-rich tasks since interaction endanger...

We address the problem of devising means for a robot to rapidly and safely learn insertion skills with just few human interventions without hand-crafted rewards or demonstrations. Our InsertionNet version 2.0 provides an improved technique robustly cope wide range use-cases featuring different shapes, colors, initial poses, etc. In particular, we present regression-based method based on multimodal input from stereo perception force, augmented contrastive learning efficient valuable features....

In recent years, industrial robots have been installed in various industries to handle advanced manufacturing and high precision tasks. However, further integration of is hampered by their limited flexibility, adaptability decision making skills compared human operators. Assembly tasks are especially challenging for since they contact-rich sensitive even small uncertainties. While reinforcement learning (RL) offers a promising framework learn control policies from scratch, its applicability...

Contact-rich assembly tasks may result in large and unpredictable forces torques when the locations of contacting parts are uncertain. The ability to correct trajectory response haptic feedback accomplish task despite location uncertainties is an important skill. We hypothesize that this skill would facilitate generalization support direct transfer from simulations real world. To reduce sample complexity, we propose learn a residual admittance policy (RAP). RAP learned movements generated by...