Robotic Limb Impact Resistance Control

In order to achieve the optimum motion characteristics of the robotic limb end after impact, the most straightforward method was to determine the conversion relationship between the motion characteristics of the joint space and the end Cartesian space. It was necessary to discover the configuration changes of the limb in real time and calculate the equivalent moment of each joint inertia. The calculated quantity of the overall process was too high. Thus, the variable damping control method based on the virtual restoring force was introduced. For the load system, it was possible to obtain its absolute movement information in relation to the Cartesian space in real time. In this case, the load could be considered as an unconstrained spatial load that was only controlled by the virtual restoring force, so as to meet the proposed requirements for impact resisting. As shown in Figure 3, the virtual restoring force acted on the mass center of the load, so that the load tended to move back to its original position. Its value varied in real time, which was related to the motion state of the load (p_t, v_t). The mapping function f_RL between the virtual restoring force and the movement status could be achieved by the Q-learning algorithm.
For the load in weightlessness, in order to reduce the deviation and bring it back to the original position, a virtual restoring force based on the spring damping model was proposed. Its virtual damping coefficient could change adaptively, as shown in Figure 3. The change between the real-time state of the load and the initial state was used as the input of the virtual restoring force, and the virtual restoring force was mainly composed of the virtual spring tension and damping force, which can be shown as follows:
where F_r represents the virtual restoring force, K is the virtual spring stiffness coefficient, D(t) is the virtual damping coefficient, X(t) is the displacement relative to the initial position after impact, and $\stackrel{.}{X}\left(t\right)$ is the velocity after impact. When the spatial load was impacted in any direction, the corresponding state changes occurred in the three-dimensional space, such as in Status B or C as shown in Figure 3. The spring damping system was applicable. That is to say, the virtual restoring force generated was always in a straight line with the displacement of the load in relation to the initial state.
For the introduced spring damping system, the corresponding impedance characteristics could be obtained by adjusting the appropriate stiffness coefficient K and damping coefficient D(t) according to the desired system characteristics. However, the fixed stiffness and damping coefficient could not simultaneously satisfy the overall impact resistance requirements. When the stiffness was fixed, if the damping coefficient was too small, the load-displacement was too large. If the damping coefficient was too large, the recovery speed after impact was too slow. Therefore, the damping coefficient was particularly critical for maximal deviation and recovery time. Considering the practical application of wearable robotic limbs, it was used to hold the handrail of the cabin to stabilize the position of the astronaut when working in a fixed spot. In this case, it was hoped that the equivalent system had a relatively large stiffness. At this time, if the method of variable stiffness was adopted, the stiffness of the system could be reduced, which was not conducive to the astronaut maintaining position. Therefore, the variable damping control method was selected in this paper. For the problem that the virtual restoring force of the fixed damping method could not fully meet the impact resistance requirements, the variable damping controller could change the virtual damping value appropriately depending on the real-time movement state, so as to meet the impact resisting requirements in different states.