As Taiwan enters into an aging society, the elderly who experience a stroke and limb dysfunction induced by cardiovascular disease and other factors face problems in maintaining daily routines, while their family and social programs encounter heavy burdens as well. The current literature on upper limb and lower limb rehabilitation training systems in the domain of rehabilitation engineering are relatively mature, but the combination of collaborative rehabilitation training of upper and lower limbs is a novel method with scant research findings at home and abroad. Therefore, in order to provide better assistance for early rehabilitation training of hemiplegic patients with a stroke, this study combines robot technology, motion intention recognition, fatigue estimation, rehabilitation evaluation, virtual reality, and adaptive interaction control of a man-machine interaction force in order to design the prototype of an exoskeleton system for collaborative rehabilitation of upper and lower limbs. The goal is to develop an exoskeleton system for coordinated rehabilitation of limbs based on virtual reality and provide a new rehabilitation method for patients. This project first develops a cooperative rehabilitation exoskeleton system for four-limb based on the theory of rehabilitation medicine. Secondly, linear extended state observer (LESO) is introduced into the design of proxy-based sliding-mode control (PSMC). Solve the problem that the original PSMC lacks stability analysis or relies on strong conjecture, and apply the PSMC based on the line expansion state observer to the four-limb collaborative rehabilitation exoskeleton system to control the body weight support system four-limb exoskeleton. Realize active safety control of cooperative passive rehabilitation of limbs. Finally, according to the experimental plan approved by the IRB, the developed system and the PSMC based on the line extension state observer were tested and verified by healthy subjects. The experimental results show that each control system operates stably and has active safety performance. In rehabilitation effectiveness evaluation, this project uses self-developed foot pressure insoles and detection software, combined with a surface electromyogram system to evaluate the effectiveness of rehabilitation training with multi-source information fusion. The limbs collaborative rehabilitation exoskeleton system developed by this project based on virtual reality has obtained the invention patent of the Republic of China (I728582). A sure foundation has been laid for the project results to enter clinical testing, and it is expected to provide better early gait rehabilitation training for patients with hemiplegia.