UR5e

From a learning point of view

The UR5e is the arm that shows up in more recent imitation-learning papers than any other, usually with a Robotiq gripper and a wrist camera. Diffusion Policy, UMI and π0 all report results on it. Control is position or velocity based through RTDE; there is no direct torque interface, so methods that need torque control (some RL fine-tuning, contact-rich learning) are easier on a Franka. The payoff is industrial reliability: the same arm a factory buys, so a policy that works here is already on production-grade hardware.