JANM-IK: Jacobian Argumented Nelder-Mead Algorithm for Inverse Kinematics and its Hardware Acceleration

Inverse kinematics is one of the core calculations in robotic applications and has strong performance requirements. Previous hardware acceleration work paid little attention to joint constraints, which can lead to computational failures. We propose a new inverse kinematics algorithm JANM-IK. It uses a hardware-friendly design, optimizes the Jacobian-based method and Nelder-Mead method, realizes the processing of joint constraints, and has a high convergence speed. We further designed its acceleration architecture to achieve high-performance computing through sufficient parallelism and hardware optimization. Finally, after experimental verification, JANM-IK can achieve a very high success rate and obtain certain performance improvements.