RoboCup ARM Challenge — Autonomous Pick-and-Place Grasping System
King's College London Group 17 entry to the RoboCup Autonomous Robot Manipulation Challenge: an end-to-end UR5e pick-and-place system in MATLAB–ROS–Gazebo combining YOLOv8 detection, PCA + ICP grasp-pose estimation, defensive inverse kinematics, and a deterministic 7-stage state machine — 100% end-to-end success on the standard object set.
Project Overview
Group robotics project (7CCEMRGP) building a fully autonomous manipulation pipeline around a simulated UR5e and Robotiq 2F-85 gripper. RGB-based detection uses an advanced YOLOv8 (BiFPN fusion + DEConv backbone, mAP 0.981 over 8 classes); 3D grasp poses come from depth-based PCA orientation classification refined by ICP template matching with a safe PCA fallback. Motion uses dual-mode planning — quintic (C²) joint-space MoveJ for transfers and SLERP Cartesian MoveL for the final approach — driven by a custom 4-seed defensive IK solver with jump protection, all orchestrated by a deterministic 7-stage finite state machine. My contribution was within the motion-planning team: the joint-space point-to-point transfer planner (developed with a teammate), early integration between YOLO perception and motion, and project coordination. Three documents are attached: the team proposal, my individual reflective report (project management, sustainability/LCA, ethics, FMEA), and the full Group 17 technical portfolio.