Environment Aware Drone Delivery
This work introduces an Offboard Non-linear Model Predictive Control (NMPC) framework implemented on the Crazyflie 2.1 hardware, with a comparative analysis against a baseline Proportional-Integral-Derivative (PID) controller for quantitative and qualitative assessment. The ACADOS solver is utilized for efficient C code generation, enabling the NMPC controller to operate at a frequency of 66.67Hz on the offboard
base station.
In the evaluation of performance metrics, a detailed analysis is conducted to discern the distinct capabilities of NMPC and PID controllers. Additionally, an exploration is undertaken to
understand the influence of payload weight on the miniature quadcopter’s performance. The findings reveal the NMPC controller's superior trajectory tracking abilities compared to its PID counterpart.
Breif Explanation of the project:(Click here for Full Report)
Introduction
The project aims to enhance the payload delivery efficiency of a quadrotor, focusing on optimizing delivery times considering the limitations of flight times and payload weights. The choice of utilizing the full non-linear dynamics of the Crazyflie underscores the commitment to achieving maximum performance.
System Modeling
To realize efficient payload delivery, the Crazyflie's state vector incorporates position, quaternion orientation, body velocity, and body angular velocity. Control input involves individual motor angular velocity. The formulation of the MPC objective function and constraints is designed to minimize deviations from a reference trajectory, considering both control inputs and system dynamics.
Controller Design
The project conducts a comparative analysis between a baseline PID controller and an advanced MPC controller. The default Crazyflie configuration uses PID, while MPC is chosen for its proven efficacy in addressing complex control challenges. The PID control architecture follows a cascaded structure involving Proportional, Integral, and Derivative terms. Meanwhile, the MPC control architecture utilizes non-linear dynamics, ROS architecture, and the ACADOS solver for optimization.
Hardware Architecture
The hardware architecture involves the integration of sensors such as the Flow Deck, Multiranger, and an Optitrack Mocap System. A crucial component is the 3D printed Mocap Marker Panel designed for accurate pose generation. The evolution of payload fabrication is discussed, starting from a lightweight design for the Flow Deck to a more flexible design with Mocap.
Trajectory Generation
Trajectory generation comprises two main categories: fixed trajectories (helical and figure 8) and trajectories incorporating obstacle avoidance. Environmental data from a motion capture system is used to generate a path plan using the Rapidly Exploring Random Trees (RRT) planner. The planning node refines the initially broad path plan into a detailed trajectory, aligning better with smaller time intervals.
Controller Implementation
Detailed insights into the implementation of both PID and MPC controllers are provided. The PID control structure involves custom trajectory development and obstacle avoidance logic. On the other hand, MPC implementation utilizes non-linear dynamics, ROS architecture, and the ACADOS solver for optimization.
Testing
The testing phase comprises eight scenarios, with four scenarios tested on each of the PID and MPC controllers. Scenarios involve custom trajectories and obstacle avoidance, with and without payload. Metrics such as Root Mean Square Error (RMSE) and Maximum Absolute Deviation (MAD) are used for evaluation.
Demo Results
The qualitative analysis of the framework is presented through visualizations of MPC tracking a custom trajectory and obstacle avoidance trajectory. The results indicate that MPC outperforms PID in terms of trajectory tracking and obstacle avoidance, emphasizing its efficiency and precision.
Conclusions
In conclusion, the project underscores the superiority of MPC over PID for trajectory tracking and obstacle avoidance. Both controllers demonstrate stability and accuracy, making them applicable in real-world scenarios, particularly in the context of payload deliveries. The findings contribute valuable insights into the field of control systems engineering.
Read Full Report here
Trajectory Obstacle Avoidance
Trajectory Figure "8"
