Investigations on Hardware Acceleration for Embedded Simulation of Multibody Dynamics

Abstract The simulation of multibody dynamics is a fundamental task in many applications, e.g., for digital twins for human-machine interaction or model-based control.Especially when deployed to embedded hardware with restricted resources, this task can quickly become a bottleneck in real-time systems.Recent research studies provide helpful insight and results using graphics processing units for multibody computations.However, those are at the current time still rarely available in embedded systems.This motivates investigating if and how hardware acceleration of multibody dynamics simulation, based on the projected Newton-Euler equations, can be achieved using as an example the Advanced Micro Devices (AMD) Zynq 7000 chip family.The Zynq 7000 platform combines a powerful microprocessor and a field programmable gate array in a single system-on-a-chip, providing a flexible environment for hardware acceleration.In this work it is introduced from a high-level point of view and a workflow for the design of hardware accelerators is described.First studies highlight the importance of parallelization and data flow through the chip.Following-up examples apply hardware acceleration to the projected Newton-Euler equations and study its effect with two selected multibody systems.Overall, depending on the system's size and complexity, a speed-up of up to a factor of 7.6 is achieved concerning the evaluation of the equations of motion.