Abstract: To improve tracking pattern control accuracy of the working parts of a hydraulic excavator robot, analysis was conducted on the working parts which were simplified into the two-dimensional robotic arm composed of arm and bucket and two joints. When establishing an inverse kinematics model, terminal position and orientation space of the bucket and joint space and cylinder space of the working parts should relate to the tracking pattern and should control the excavator robot in each space. To improve precision in tracking the desired pattern, two adaptive neural-fuzzy inference systems (ANFIS) were used to determine inverse mapping relations between the x and y joint coordinates and joint angle, and an ANFIS inverse mapping model was established. I/O curve data of inverse mapping was selected to train ANFIS structure, with an I/O mapping curve of a fuzzy model used to obtain a corresponding joint angle based on a given desired excavation trace. Finally, the proposed fuzzy model was used to trace an expected motion pattern, and simulation results showed that tracking precision met actual demands.