Autonomous mobile robots (AMRs) have been proven useful for smart factories and have the potential to revolutionize critical missions, such as disaster rescue. AMRs can perceive the environment, plan for assigned tasks, and act on the plan. Motion control is critical to the robot's action, which is accomplished through trajectory optimization to refine the robot's states using a physics model. However, the high computational complexity of trajectory optimization poses significant challenges for AMRs with limited power and computing resources. In addition, the computations of trajectory optimization are adapted to the physical configurations of AMRs, requiring the underlying computing resources to be flexible enough while maintaining high efficiency.

To address these challenges, this work proposes a prototype of motion-control SoC that perform low-latency, energy-efficient trajectory optimization while supporting various AMR configurations. The SoC includes an ARM Cortex-M3 MCU, a trajectory optimization accelerator, and a post-processing unit. The trajectory optimization accelerator leverages parallelism in the trajectory optimization algorithm. It contains a 4x4 array of processing elements that compute in parallel to reduce the trajectory optimization latency. The ARM Cortex-M3 MCU is programmed by software for flexible system configurations, scheduling, and auxiliary data processing.

The SoC is implemented through the cell-based ASIC design flow. The ARM Cortex-M3 MCU and the AHB bus infrastructure are synthesized using the open-source RTL implementation, and other parts of the SoC are designed in-house. The software programs for the MCU are compiled by the ARM GNU toolchain. Fabricated in 28nm CMOS technology, the motion-control SoC integrates 5.3M logic gates and 146KB on-chip SRAMs in a core area of 3.56mm². The SoC dissipates 142mW at a 200MHz clock frequency from a 1.0V supply. The chip achieves a 4.935kHz maximum control rate for a 7-Degree-of-Freedom robot arm. Compared to the state-of-the-art motion-control accelerator, this work achieves a 22× improvement in the maximum control rate. It also delivers 350× higher energy efficiency, despite the added flexibility. This prototype demonstrates a promising solution for future agile humanoid robots that require ultra-fast response and robust control.