Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Design and Implementation of Automatic Assisted Aiming System For Robomaster EP Based on YOLOv5

Junjia Qin, Kangli Xu

TL;DR

This work addresses autonomous target recognition and accurate aiming for the Robomaster EP by integrating a YOLOv5-based detector with DeepSORT tracking, Kalman filtering, and a PID controller enhanced with FIR for gimbal control. It introduces the AAAS-2021 dataset and demonstrates end-to-end development from data collection to deployment, including architecture optimizations via Strided-Convolution and edge-device evaluation. The system combines real-time detection, robust multi-target tracking, and trajectory/deviation compensation to improve targeting accuracy and competitive performance in autonomous RoboMaster scenarios. The proposed framework offers a practical, extensible approach for autonomous aiming in high-speed robotic combat and related real-time targeting tasks.

Abstract

In the crucial stages of the Robomaster Youth Championship, the Robomaster EP Robot must operate exclusively on autonomous algorithms to remain competitive. Target recognition and automatic assisted aiming are indispensable for the EP robot. In this study, we use YOLOv5 for multi-object detection to identify the Robomaster EP Robot and its armor. Additionally, we integrate the DeepSORT algorithm for vehicle identification and tracking. As a result, we introduce a refined YOLOv5-based system that allows the robot to recognize and aim at multiple targets simultaneously. To ensure precise tracking, we use a PID controller with Feedforward Enhancement and an FIR controller paired with a Kalman filter. This setup enables quick gimbal movement towards the target and predicts its next position, optimizing potential damage during motion. Our proposed system enhances the robot's accuracy in targeting armor, improving its competitive performance.

Design and Implementation of Automatic Assisted Aiming System For Robomaster EP Based on YOLOv5

TL;DR

This work addresses autonomous target recognition and accurate aiming for the Robomaster EP by integrating a YOLOv5-based detector with DeepSORT tracking, Kalman filtering, and a PID controller enhanced with FIR for gimbal control. It introduces the AAAS-2021 dataset and demonstrates end-to-end development from data collection to deployment, including architecture optimizations via Strided-Convolution and edge-device evaluation. The system combines real-time detection, robust multi-target tracking, and trajectory/deviation compensation to improve targeting accuracy and competitive performance in autonomous RoboMaster scenarios. The proposed framework offers a practical, extensible approach for autonomous aiming in high-speed robotic combat and related real-time targeting tasks.

Abstract

In the crucial stages of the Robomaster Youth Championship, the Robomaster EP Robot must operate exclusively on autonomous algorithms to remain competitive. Target recognition and automatic assisted aiming are indispensable for the EP robot. In this study, we use YOLOv5 for multi-object detection to identify the Robomaster EP Robot and its armor. Additionally, we integrate the DeepSORT algorithm for vehicle identification and tracking. As a result, we introduce a refined YOLOv5-based system that allows the robot to recognize and aim at multiple targets simultaneously. To ensure precise tracking, we use a PID controller with Feedforward Enhancement and an FIR controller paired with a Kalman filter. This setup enables quick gimbal movement towards the target and predicts its next position, optimizing potential damage during motion. Our proposed system enhances the robot's accuracy in targeting armor, improving its competitive performance.
Paper Structure (23 sections, 21 equations, 3 figures, 3 tables, 2 algorithms)

This paper contains 23 sections, 21 equations, 3 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Programming Task
  4. Experiment 1: End-to-End Model Development: From Dataset Construction to Deployment
  5. Dataset Construction
  6. Pre-Trained Configuration
  7. Train Model
  8. box_loss
  9. obj_loss
  10. cls_loss
  11. metrics_precision
  12. metrics_recall
  13. mAP (mean Average Precision)
  14. YOLOv5n Model Optimization
  15. Model Deploy
  16. ...and 8 more sections

Figures (3)

  • Figure 1: Recognizable objects.
  • Figure 2: Regression parameters for each algorithm. Red represents the actual parameters of the water pellet falling.
  • Figure 3: Reflects the aiming status of the Automatic Assisted Aiming System, A sequence of four images representing the progression of states over time, from left (earliest) to right (latest).