Aggressive Trajectory Tracking for Nano Quadrotors Using Embedded Nonlinear Model Predictive Control

Muhammad Kazim,Hyunjae Sim,Gihun Shin,Hwancheol Hwang,Kwang-Ki K. Kim
2023-12-02
Abstract:This paper presents an aggressive trajectory tracking method for a small lightweight nano-quadrotor using nonlinear model predictive control (NMPC) based on acados. Controlling a nano quadrotor for accurate trajectory tracking at high speed in dynamic environments is challenging due to complex aerodynamic forces that introduce significant disturbances and large positional tracking errors. These aerodynamic effects are difficult to be identified and require feedback control that compensates for them in real time. NMPC allows the nano-quadrotor to control its motion in real time based on onboard sensor measurements, making it well-suited for tasks such as aggressive maneuvers and navigation in complex and dynamic environments. The software package acados enables the implementation of the NMPC algorithm on embedded systems, which is particularly important for nano-quadrotor due to its limited computational resources. Our autonomous navigation system is developed based on an AI-deck that is a GAP8-based parallel ultra-low power computing platform with onboard sensors of a multi-ranger deck and a flow deck. The proposed method of NMPC-based trajectory tracking control is tested in simulation and the results demonstrate its effectiveness in trajectory tracking while considering the dynamic environments. It is also tested on a real nano quadrotor hardware, 27-g Crazyflie 2.1, with a customized MCU running embedded NMPC, in which accurate trajectory tracking results are achieved in dynamic real-world environments.
Robotics
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to develop an efficient and accurate control method for the nano - quadrotor drone (Crazyflie2.1), enabling it to accurately track the desired trajectory in a dynamic environment while considering the system's dynamics and constraints. Specifically, the paper focuses on how to achieve high - precision and efficient trajectory - tracking control, allowing the nano - quadrotor drone to perform high - speed and high - acceleration maneuvers in complex environments. ### Problem Statement The research aims to develop a highly accurate and efficient control method for the embedded nano - quadrotor drone Crazyflie2.1 to accurately track the desired trajectory while considering the system's dynamics and constraints. This method needs to be able to handle aggressive maneuvers with high precision and fast response times and navigate in a dynamic environment. ### Challenges 1. **Accurate Modeling**: Accurately model the dynamics of the nano - quadrotor drone. 2. **Constraint Handling**: Handle various constraints in the system. 3. **Real - Time Implementation**: Implement real - time control on an embedded system with limited resources. ### Solutions To address these challenges, the research proposes using nonlinear model predictive control (NMPC) as the control strategy and using the acados toolbox to implement the NMPC algorithm on the embedded system of the nano - quadrotor drone. The specific steps are as follows: 1. **System Modeling**: Establish a dynamic model of the nano - quadrotor drone, including translational and rotational motions. 2. **Optimization Problem**: Define an optimization problem, and by solving this problem, determine the optimal control input to enable the drone to track the desired trajectory. 3. **Real - Time Solution**: Use the high - performance solvers provided by acados (such as HPIPM and SQP RTI) to solve the optimization problem in a real - time environment. 4. **Experimental Verification**: Verify the effectiveness of the proposed method through simulation and actual hardware experiments. ### Main Contributions 1. **Efficient Implementation**: Implement the NMPC control algorithm on an embedded platform with limited resources. 2. **High - Precision Tracking**: Achieve high - precision trajectory tracking in a dynamic environment. 3. **Real - Time Performance**: Through the optimization algorithm and high - performance solvers, achieve real - time control. ### Future Work 1. **Multi - Drone Extension**: Extend this method to multiple nano - quadrotor drones. 2. **Application Expansion**: Apply this method to other unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs). 3. **Integration of Other Technologies**: Combine reinforcement learning and human - machine collaborative control technologies to improve the safety and reliability of the system. Through these efforts, this research provides a new solution for the precise control of nano - quadrotor drones in complex environments.