Residual policy in the robot's control software

The robot's controller now runs the residual policy, both networks and the behaviour inputs as trained. In the ROS 2 simulation the default walk climbs the 3 cm stairs; the 4 Hz gait does so only once the simulator's joint torque limit is set from the motor's joint-side figures.

• Controller output against the policy computed offline: equal within 1.2e-6 on 484 recorded steps, inference 0.5 ms per step
• 3 cm stairs in the ROS 2 simulation: default walk crosses (0.254 m on the top tread), stand command holds still
• 4 Hz gait on the 3 cm stairs: 0 of 3 with the simulator's 1.37 N·m joint limit, 2 of 3 with the motor's joint-side figures (3 N·m, falling to 0 at 38 rad/s)
• The simulator's default limit is left as it was; changing it is open