Static and Dynamic Dispatch in Robotics Control Loops: A Practical Performance Study in Rust

Le résumé fourni par la source

Robotic control loops require tight timing guarantees while remaining maintainable as hardware and platforms evolve. In Rust, trait-object dispatch enables open-world composition of heterogeneous components, whereas static dispatch (e.g., generics or enum-based polymorphism) can reduce call overhead but often introduces closed-world designs and boilerplate. We benchmark representative dispatch mechanisms in Rust and contextualize them against end-to-end actuator communication latency on a real robotic platform. In our setup, blocking device communication dominates loop time (hundreds of$\mu \mathrm{s}$to a few ms), while dispatch overhead remains in the tens of ns—an orders-of-magnitude separation that renders dispatch choice irrelevant to end-to-end timing in I/O-bound loops. In microbenchmarks without I/O, the fastest static mechanisms exhibit lower latency (approximately$\text{5 0 \%}$in our measurements, corresponding to$\sim \text{1. 5 ~ n s}$) and more predictable execution than trait-object dispatch, making them preferable in compute-bound hot paths and extremely high-rate inner loops. As actuator and bus latencies decrease and communication is moved off the critical path (e.g., via asynchronous I/O), such software-level costs can consume a larger fraction of the real-time budget. Overall, our results support using dynamic dispatch at hardware abstraction and composition boundaries today, while motivating static dispatch in performance-critical kernels as robotic systems become increasingly compute-bound.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.