Bio-inspired walking: A FPGA multicore system for a legged robot

Abstract

Previous legged robots use single or multi-microcontroller systems to control their motions. This work is a complete robot control system, implemented as a Multi-Processor System-on-Chip (MPSoC), on a Spartan 3A Field Programmable Gate Array (FPGA). Novel features of this system include encapsulation of the various levels of control, low communication latency between processors (4 clock cycles at 50 MHz), and ease of use for the control system researchers. The MPSoC implementation combines the performance benefits of processing control loops in parallel, with the size and mass advantages of a single IC solution. The system comprises one soft processor that is used for high-level decisions regarding the robot's overall movements, and six soft processors that run independent, low-level control loops for each of the six legs. The low-level control loop frequency can reach up to 2 kHz, and is only limited by the Analog to Digital Converter (ADC) sample rate. Coordination between legs occurs at 100 Hz. This design uses 90% of the user I/Os, 57% of the flip flops, 70% of the LUTs, 18% of the DSPs and 89% of the block RAMs on the FPGA, with a system operating frequency of 50 MHz. A legged robot, Abigaille-III, uses this control system to walk on flat and uneven surfaces.