Low-cost adaptive obstacle avoidance trajectory control for express delivery drone

Introduction

The paper discusses the obstacle avoidance trajectory control (OATC) problem for uncrewed aerial vehicles (UAVs), which is important for applications like exploration, swarm flight, and coronavirus prevention. Since drones are not limited by infrastructure, they can lower costs and delivery times. However, varied quadcopter parameters make controller transplant difficult. This paper proposes a NLVG-PID controller that can reduce overshoot in fixed-gain PID controllers. The key contributions are:

1. Analysis of the fixed-gain PID controller structure and quadcopter dynamics/kinematics models.

2. Design of a NLVG-PID controller for a delivery quadcopter, using extremum seeking to learn optimal nonlinear PID parameters. This is the first NLVG-PID obstacle avoidance controller for delivery drones.

3. Simulation verification of the proposed control scheme on a quadcopter under two collision-free trajectories.

The paper also describes the quadcopter control system scheme in Fig. 1.

The paper describes the key sections as follows:

Section 2 details the mathematical model of the quadcopter's dynamics, kinematics, and obstacle detection pipeline.

Section 3 develops the NLVG-PID control framework and an Evolutionary Strategies method for optimal parameter tuning. It also provides a stability analysis of the closed-loop system.

Section 4 verifies the quadcopter's NLVG-PID controller through testing in various flight scenarios, including four-side route, 8-character route, and scenic spiral position tracking.

Section 5 concludes the work.

Problem formulation

The paper assumes the structure of the express delivery drone is known, but the specific load weight and precise drone model are unknown. Obstacle avoidance is performed in three-dimensional space. First, the threat area is circled in the top view, as shown in Fig.2. The drone then avoids obstacles and chooses to fly upward or downward based on the Euclidean distance cost.

The text describes a coordinate system used to analyze the rotation and translation of a UAV. The body-fixed coordinate system (B-frame) is represented as B=(XB,YB,ZB), and the earth inertial frame (E-frame) is represented as E=(XE,YE,ZE). The velocity of the quadcopter in the E-frame is represented as VE=(vex,vey,vez). The schematic diagram of the quadcopter control system, as shown in Fig. 1, includes the components described in the text.

Controller design

The text provides a detailed explanation of a nonlinear variable gain PID (NLVG-PID) controller for quadcopter control. The key points are:

1. The standard PID controller is defined, with proportional, integral, and derivative terms. This is then extended to a nonlinear formulation, where the gains are varied based on the size of the error.

2. Smaller errors use lower gain values, while larger errors use higher gain values. This helps improve the controller's fast response capability for different error magnitudes.

3. The nonlinear gain functions are defined mathematically, with different thresholds and gain values for the P, I, and D terms.

4. An extremum-seeking (ES) optimization approach is used to automatically tune the PID gain parameters online. This iteratively updates the gains to minimize a cost function related to tracking error.

5. The paper also discusses planning obstacle avoidance trajectories for the quadcopter using sensor information and geometric constraints. This ensures safe, collision-free flight.

Overall, the NLVG-PID controller with ES tuning and obstacle avoidance strategy provides an advanced control framework for improved quadcopter performance and safety.

Simulation and results

The paper describes a simulation experiment conducted to verify the efficiency of the NLVG-PID controller. The experiment focused on the internal loop angle control of a quadcopter.

The test compared the performance of a standard PID controller against the NLVG-PID controller, which uses three cost functions: IAE (integrated absolute error), ITAE (integration time and absolute error), and ITSE (integrated time-weighted squared error).

The results show the NLVG-PID controller had a peak angle response time of 0.15 seconds and a peak position response time of 0.37 seconds. These performance metrics are presented in Fig. 5 and Table 1 of the paper.

The key points from the provided text are:

1. The initial parameters for the quadcopter's inner loop controller are set as:
   • Roll (φ) PID: (8, 0.1, 5)
   • Pitch (θ) PID: (8, 0.1, 5)
   • Yaw (ψ) PID: (8, 0.1, 5)
   • The outer loop position controller initial parameters are similar but with slightly lower values.
2. The scalar gains δ1 and δ2 are set to 0.01 and 0.838, respectively, and the control parameters are learned using Evolutionary Strategies.
3. The simulation uses a step size of 0.01 seconds and a total time of 140 seconds.

4.1 Example 1: Storm Path Following

• The aircraft follows a storm-type obstacle avoidance path, which is a continuous arc with increasing radius and height.
• Both the PID and NLVG-PID controllers are able to reduce the attitude angle and position errors as the aircraft executes the storm path.
• The NLVG-PID controller shows improved performance compared to the fixed-gain PID.

4.2 Example 2: Lissajous Curve Following

• The aircraft tracks a 3D Lissajous curve, which has periodic changes in height and position.
• Similar to the storm path, the NLVG-PID controller outperforms the fixed-gain PID in reducing attitude angle and position errors.
• However, the performance improvement is slightly reduced compared to the storm path example, likely due to the increased complexity of the Lissajous trajectory.

Conclusion

This paper proposes a new design scheme for adaptive learning control of quadcopter flight controllers. The authors present an NLVG-PID controller for the UAV model and use Extremum Seeking (ES) to learn optimal NLVG-PID parameters offline. Numerical simulations show the proposed adaptive learning controller can reduce response overshoot and settling time in typical 3D flight paths, such as storm paths.

References