Active Balance System for a Humanoid Robot

Exhibitor: Toby Low

Supervisor: Gordon Wyeth

Research Group: Complex and Intelligent Systems

Industry Sector: Scientific / Research Services

The main problem in the development of humanoid robots is in implementing or mimicking the Central Nervous System (CNS), brain and the sensors of the human body. With these tools, humans are able to balance and perform movements subconsciously. In order for a humanoid robot to perform these trivial tasks, many control loops must be formulated to keep the robot stable and perform the desired tasks correctly.

The scope and goal of this thesis is to design and implement an active balance system that will enable the robot to remain upright despite adverse disturbances. These disturbances were classified into 3 categories: Small disturbances, slight changes in the walking surface, small forces and errors in joint positions. Medium disturbances, medium changes in ground surface, forces that will tip the robot up to 20-30 degree angle. Large disturbances, large forces where the robot is required to move legs or take a step counter force and remain upright. Due to the complexity involved for countering large disturbances, the scope of the thesis was limited to countering small to medium disturbances. Also to allow modularity between various control systems being developed for GuRoo, the balance system was limited to the control of the torso and upper body limbs only.

To counter these small to medium disturbances, a inertial measurement unit or IMU sensor system to detect dynamic disturbances was implemented and a simplified model of the GuRoo robot was formed shown in Figure 1. A control system to keep the posture on the robot upright was designed using this model and was then simulated using Dynamechs. The control system was then tweaked accordingly and tested with small to medium forces. The balance system was then implemented on the real robot.

Figure 1: Left - IMU Sensor. Right - GuRoo Balance Model, Two Mass Inverted Pendulum.

The designed control system was implemented successfully on the humanoid robot keeping the robot's overall angle close to the desired angle. Although this was successful, the implications of trying to keep the robot upright were not foreseen and thus medium size forces caused the robot to oscillate and fall over. The active balance system was also implemented on top of a previous gait (walk) and found to be successful in helping the robot remain balanced despite small forces.

Many issues still need to be tackled in future active balance systems whereby more complex control techniques and behaviours will need to be designed to counter medium to large forces. Also control of other joints of the robot such as knees, ankles and hips will play a large part in trying to balance the robot. Foot sensors were also missing from the robot but are currently being developed which will complete the proposed overall balance sensor system and will be used in future active balance systems to come.

If you are interested in finding more in depth information, read the thesis document attached below.

Thesis Document (PDF)

Other Related Files