Nowadays, doctors combine lots of sources of information to assess a patient's walking problems (i.e. physical examination, clinical acquisition) to propose the correct surgery. Such information would be sufficient for a surgeon to predict the effect of a specific surgical operation on the patient's gait. Often, due to the complexity of such examinations and diagnostic data, it can be challenging to accomplish an accurate prediction for the result of invasive surgery. This is not surprising considering that the musculoskeletal (MSK) system of the human body is a highly dimensional complex dynamic system. MSK is comprised of anatomical factors, such as bone geometry, muscle conditions, bodyweight; factors that account for anomalies in the performance of the human walking movement. Corrective surgery aims to alter anatomical conditions that impact negatively the overall motion in a unique way for each patient. Pathological gait patterns can be divided into either neuromuscular or musculoskeletal etiologies Some gait abnormalities may result from structural deformities of a bone, joint, soft tissue, or neuromuscular condition. The main focus of this master thesis is to model and simulate pathological gait resulting from motor impairments using machine learning techniques. Reinforcement learning (RL) has recently demonstrated its capabilities for controlling simulated human bodies and reproducing accurately the kinematics of human motion. However, the performance of the RL agent is directly determined by the structure of the musculoskeletal body and its actuator type (i.e torques or muscles). Modeling human pathological gaits can support doctors in the diagnostic process about the impact of various types of surgical operations on MSK that influence the patient's gait. The desirable impact of a successful simulation and prediction would be to reduce the amount of engagement between a patient and a surgeon and ultimately increase the percentage of effective surgical gait operations. With the use of real patient kinematics composed by the Laboratory of Kinesiology of Hôpitaux Universitaires de Genève (HUG) and the modification of Opensim simulator, provided by the BioRob team from EPFL. We try to develop applications using reinforcement and imitation learning methods to derive policies in order to predict a patient gait from the given kinematics. This trained controller allows surgeons for a wide range of analyses on virtual surgeries and their remedial effects on bringing a pathological gait closer to a healthy gait.