An EtherCAT-Based Real-Time Control System Architecture for Humanoid Robots

The design of humanoid robots naturally requires the simultaneous control of a high number of joints. Moreover, the performance of the overall robot is strongly determined by the low-level control system as all high-level software e.g. for locomotion planning and control is built on top of it. In order to achieve high update rates and high bandwidth for the joint control, an advanced real-time control system architecture is required. However, outdated communication protocols with associated limits in the achievable update rates are still used in nowadays humanoid robots. Moreover, the performance of the low-level control systems is not analyzed in detail or the systems rely on specialized hardware, which lacks reliability and persistence. We present a reliable and high-performance control system architecture for humanoid robots based on the ETHERCAT technology. To the authors' knowledge this is the only system, which operates at control rates beyond 2 khz and input/output latencies below 1 ms. Our control architecture includes a learning-based feedforward control strategy to improve joint tracking performance. The improved joint control method and the communication system are evaluated on our humanoid robot LOLA. Our software framework is available online to allow other researchers to benefit from our experiences.