Improving Dynamics Estimations and Low Level Torque Control Through Inertial Sensing

In 1996, professors J. Edward Colgate and Michael Peshkin invented the cobots as robotic equipment safe enough for interacting with human workers. Twenty years later, collaborative robots are highly demanded in the packaging industry, and have already been massively adopted by companies facing issues for meeting customer demands. Meantime, cobots are still making they way into environments where value-added tasks require more complex interactions between robots and human operators. For other applications like a rescue mission in a disaster scenario, robots have to deal with highly dynamic environments and uneven terrains. All these applications require robust, fine and fast control of the interaction forces, specially in the case of locomotion on uneven terrains in an environment where unexpected events can occur. Such interaction forces can only be modulated through the control of joint internal torques in the case of under-actuated systems which is typically the case of mobile robots. For that purpose, an efficient low level joint torque control is one of the critical requirements, and motivated the research presented here. This thesis addresses a thorough model analysis of a typical low level joint actuation sub-system, powered by a Brushless DC motor and suitable for torque control. It then proposes procedure improvements in the identification of model parameters, particularly challenging in the case of coupled joints, in view of improving their control. Along with these procedures, it proposes novel methods for the calibration of inertial sensors, as well as the use of such sensors in the estimation of joint torques.