Robotic Insertion

Robotic insertion concerns placing parts such as pegs, connectors, and flexible rods through openings, as well as guiding needles or instruments into tissue, where small alignment errors and contact can cause failure or damage. Methods combine vision, tactile and force sensing with compliant or hybrid position-force control; learning approaches include imitation learning, reinforcement learning, diffusion policies, and predictive world models, often trained or tested in simulation before real-world deployment. Benchmarks range from peg-in-hole tasks with varied shapes and clearances to connector assembly, flexible-object insertion, and image-guided medical procedures. A central challenge is achieving reliable generalization to unseen geometries and uncertain contact while keeping forces safe, with sim-to-real transfer and comparable evaluation still important open issues.