Determining and Reducing the Likelihood of Falling

The state of the art in diagnosing unsafe behavior relies on a combination of human expertise or exhaustive testing. For example, many existing tools  to estimate the likelihood that a locomoting system (human or robotic) will fall rely on heuristic measures extracted from physician experience or on perturbing a steady state gait at arbitrary instances in time. As a result, automated tools for predicting falls have not emerged. 
Given a model, we have developed optimization tools to explicitly measure the amount of disturbance that a specific gait can tolerate at arbitrary instances during a steady state locomotor pattern. Current work, is focused on improving the scalability of this approach and validating its utility via experimentation.


  1. R. Vasudevan, “Numerical Advection of Probability Densities for High Dimensional Systems,” in SIAM Conference on Applications of Dynamical Systems, 2015. 
  2. A. Majumdar, R. Vasudevan, M. M. Tobenkin and R. Tedrake, “Convex Optimization of Nonlinear Feedback Controllers via Occupation Measures,” International Journal of Robotics Research, vol. 33, pp. 1209-1230, August 2014. [url]
  3. V. Shia, R. Vasudevan, R. Bajcsy and R. Tedrake, “Convex Computation of the Reachable Set for Controlled Polynomial Hybrid Systems,” in IEEE Conference on Decision and Control, 2014. [pdf]