Code Generation for Conic Model-Predictive Control on Microcontrollers with TinyMPC

锥约束在许多重要的控制应用中都会出现，例如腿部运动、机器人操纵和自主火箭着陆。然而，目前的锥优化问题求解器在浮点运算和内存占用方面都有相对较大的计算需求，使它们在小型嵌入式设备上的使用变得不切实际。我们将针对低功耗嵌入式控制应用的开源高速求解器TinyMPC扩展到处理二阶锥约束。我们还提供了代码生成软件，以便在各种微控制器上部署TinyMPC。我们对我们生成的代码进行基准测试，与最先进的嵌入式QP和SOCP求解器进行比较，证明了TinyMPC相对于ECOS的速度提高了两个数量级，同时内存消耗更少。最后，我们展示了TinyMPC在Crazyflie上的有效性，Crazyflie是一个具有快速动力学的轻量级资源受限四旋翼飞行器。TinyMPC及其代码生成工具可在https://tinympc.org上公开获取。

TinyMPC: A High-Speed Solver for Conic Optimization on Resource-Constrained Embedded Systems

The paper presents an extension of TinyMPC, an open-source solver for low-power embedded control applications, to handle second-order cone constraints. The authors also provide code-generation software to deploy TinyMPC on microcontrollers. They benchmark their generated code against state-of-the-art embedded QP and SOCP solvers, demonstrating a two-order-of-magnitude speed increase over ECOS while consuming less memory.

The paper demonstrates the efficacy of TinyMPC on a resource-constrained quadrotor with fast dynamics, the Crazyflie. The code-generation tools are publicly available at https://tinympc.org. The paper also highlights the heavy computational demands of current solvers for conic optimization problems and the impracticality of using them on small embedded devices.

Related work includes embedded QP and SOCP solvers, such as ECOS and OSQP, as well as other optimization techniques for resource-constrained systems, such as model predictive control and reinforcement learning-based control.