Code and Data to perform trajectories optimization from bicycle (+rider) motion and get the control troques of the rider.
- The torque based cyclist balance strategy is poorly describe in the litrature, need for observation, quantification and description
- The link between cyclist control actions and workload is still unknown, gaze based workload metrics and balance performance metrics can be used as workload proxy
This code is meant to be a open tool to estimate cyclist's control torques from bicycle motion. Sources of motion can be diverse (mocap, IMU, GPS etc.). Complexity of the model can be adjusted : rigid rider, leaning rider, more advanced...
- impact of model complexity on tracking performance and accuracy (real data + synthetic data)
- What is a/the good/best performance metric for this method?
- sensitivity analysis to bicycle and rider parameters
- IMU to torques ? What's the best setting ?
- compare torque control metrics to gaze metrics
- make a rigid rider model -> Done
- generate synthetic data with a leaning rider, ie: we know the torques and the motion and we test our method
- run the method on real data :
- hands-off cycling, do we get a very low steer torque? -> We get full steer torque, most controllabe input
- data with steer torque sensor, compute accuracy
- data with chest cast, do we get a very low roll torque?
- find correlations between accuracy and tracking performance
- run Morris analysis with bicycle parameters