Design of a control system for a diseased pig carcass transport robot based on laser slam and machine vision

Abstract

• An autonomous robot system is developed for carcass handling in ASF prevention. • YOLO detection and RGB-D sensing are used to locate carcasses in 3D space. • Navigation uses 3D LiDAR-based obstacle avoidance and path planning. • A scooping-grasping end-effector enables contact-free carcass removal. • Field tests show >80 % success rate and 2.9–4.3 min task time in pig barns. In the context of African swine fever (ASF) prevention, the safe removal and transport of carcasses constitutes a critical biosecurity measure. Traditional manual handling and semi-automated carts still rely heavily on human intervention, resulting in low efficiency and a high risk of cross-infection. To address these challenges, this study proposes an autonomous carcass-handling robot system designed to eliminate direct human contact and enhance handling efficiency. The system introduces a novel perception–navigation integration framework that combines YOLO-based object detection, keypoint recognition, and RGB-D sensing to not only identify carcasses but also automatically generate navigation target poses from detection results, enabling truly end-to-end autonomous scooping and transport. This perception module is seamlessly integrated with a 3D LiDAR-based SLAM and path planning system, while a front-mounted scooping–grasping end-effector is custom-designed for carcass retrieval in confined pig barn environments. A lightweight coordinate conversion method translates perception outputs into actionable navigation commands with minimal computational overhead. Both hardware and software subsystems have been fully implemented, and multi-distance field trials have been conducted in a commercial pig barn. The results indicate that the system achieves up to 95 % success in individual task stages and maintains an overall carcass-handling success rate above 80 %. Average task completion time ranged from 2.9 to 4.3 min across different distances. Compared with conventional laser SLAM–vision combinations, the proposed approach delivers higher autonomy, stronger adaptability to varying layouts and lighting conditions, and reduced computational cost, demonstrating considerable potential for practical implementation in biosecure livestock farming environments.