Modular Neural Controllers for Predictive Simulations of Lower Limb Tasks

The loss of locomotion and the ability to maintain posture potentially endangers our personal independence and quality of life. The causes of disabilities in the sensorimotor systems might be incidental while others, like ageing, are inevitable for all of us. However, the underlying principles that rule the control of posture and locomotion and their relation to techniques and devices for rehabilitation are still to be revealed. Predictive Neuromuscular Modelling (PNM) can bridge the gaps in this knowledge and provide information about neurological and muscular activity that are difficulty accessible by current tools of measurement. While speculative, this method can be used to test hypothesis about the interplay between the neural and the musculoskeletal system, and the environment. PNM accelerates research by identifying experimental conditions based on optimality principles, while facilitates the assessment of devices such as orthosis or prothesis, or human-in-the-loop robots. This thesis investigates the implications of reflex-based control and modularity in human motion through PNM. I implemented two neural controllers, one for Sit-To-Stand (STS), and other for Stand-To-Walk (STW). For STS I created two versions of the same controller which vestibular and muscle-length reflexes are structured according to the typical kinematic division of the STS cycle (4-phases controller) or according to the critical event of lift-off (2-phases controller). Both controllers achieved replicating human-like STS motion. Nonetheless, 2-phases controller presents motions which temporal features are consistent with experimental data and which convergence during optimization is remarkable better when compared to the 4-phases controller. Thus, the motor control underlying STS could be simpler than the biomechanical descriptions associated with this motion. Also, the results indicate that the same sensorimotor processes can be used consistently for different motor subtasks. The Modular Posture and Locomotion (MPL) model further exploits this last concept. I developed a modular architecture where muscle synergies are recombined in motor-oriented neural pathways according to the commands of internal models and supraspinal centres. The MPL model is tested in a STW transition. The model reorganizes muscle synergies in standing and walking tasks and shows a gait pattern which is consistent to experimental data. The model is also capable to switch to slower and faster walking once steady gait is achieved. These features indicate that the MPL model can be a first step to build unified architectures for neuromuscular modelling, which can overcome the typical limitations of previous models, such as discretization of the motion skill space or multidimensional sets of solutions for a motion.