Neural network balance control of hopping robots in flight phase under unknown dynamics

In this paper, a balance control scheme is introduced for hopping robots using neural networks. The control strategy uses a trade-off strategy to achieve both hip and body angle control simultaneously with a single controller, which yields reduced complexity. Hence, the proposed controller allows the hopping robot to track a pre-defined state trajectory in flight-phase despite its modeling uncertainties. The overall dynamics complexity is decreased so that a conventional PID controller can be used for leg extension´s length tracking. Unlike other control techniques, no velocity sensor, gyroscope, camera, or body location estimation is needed. Furthermore, no offline learning or a priori system dynamics knowledge is required. Results for different situations highlight the performance of the proposed control approach in compensating for nonlinearities and disturbances.