Continuous and Unified Modeling of Joint Kinematics for Multiple Activities

Intuitive control of powered prosthetic lower limbs is still an open-ended research goal. Current controllers employ discrete locomotion modes for well-defined and frequently encountered scenarios such as stair ascent, descent, or ramps. Non-standard movements side-shuffling into cars avoiding obstacles are challenging to limb users. Human a continuous motion comprising rhythmic non-rhythmic movements, fluidly adapting the environment. It exhibits strong inter-joint coordination movement single joint can be largely predicted based on rest body. We explore unified kinematics estimation strategy wide variety without need labeled examples. Our data-driven approach uses natural body from intact trunk generate kinematic reference trajectory joints. Wearable sensors were worn by 63 subjects disabilities record full-body during typical (flat ground stairs), atypical (side shuffles, weaving through cones, backward walking). A Recurrent Neural Network (RNN) was trained predict right ankle knee other joints inputs. Results assessed 3 different test previously unseen network. All predictions had RMSE less than 7.5 degrees high correlation across activities. These offline robust subject-specific variations walking speed step length. Additionally, feasibility using towards in real-time, systems designed with participant. The controller acquired live kinematics, generated pre-trained neural network, demonstrated capability actuate prosthesis treadmill task.