This study presents design and human-robot collaboration framework for a reconfigurable supernumerary robotic limb (SRL) system, focusing on obstacle-avoidance trajectory planning and adaptive control strategies. The SRL features a modular, lightweight rigid structure with five rotational and one translational degree of freedom, and it is mounted on the operator’s waist to enhance stability. To address interference risks during overhead work, we propose an RRT*-based trajectory planning algorithm combined with cubic B-spline smoothing, which optimizes path length and minimizes vibrations while ensuring collision-free motion. For human-robot coordination, a multi-modal control framework integrates inertial measurement units (IMUs), electromyography (EMG), and motion cameras to detect user intent with a finite state machine (FSM) model. The system employs reinforcement learning-tuned variable impedance control, adapting stiffness and damping parameters in real-time to improve precision and safety. Simulations demonstrate the SRL’s ability to navigate static obstacles, achieving smooth trajectories via B-spline interpolation.