Aggressive Flight Performance using Robust Experience-driven Predictive Control Strategies: Experimentation and Analysis

Mosam Dabhi, Alexander Spitzer and Nathan Michael
Tech. Report, CMU-RI-TR-19-08, June, 2019

Download Publication

Copyright notice: This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.


Due to a plethora of applications of resource-constrained Micro Air Vehicles (MAV), there has been an increasing demand to fly aggressively high-speed flights to accomplish the tasks in the minimum amount of time for maintaining the resources. However, as the MAV executes these high-speed behaviors, the safety of the vehicle is often compromised in order to achieve the desired acceleration and aggressive behaviors. We enhance Robust Experience-driven Predictive Control (R-EPC) strategy with the ability to track aggressive a priori specified trajectories perfectly. This is achieved by appropriately handling higher order derivatives of the provided trajectories and appropriately formulating the Quadratic Program optimization problem to account for the higher order derivatives. Implications of leveraging full state dynamics of the MAV in a single control loop over separated cascaded control loop is also considered for situations where the feedforward terms are not readily available.

R-EPC generates a set of parameterized controllers for specific scenarios which can be stored and re-used to reduce the expensive online optimization costs onboard a computationally constrained platform, while the robust behavior tightens those constraint bounds to induce a conservative behavior in the presence of uncertain state estimates to further enforce safety. We further increase the efficiency of R-EPC to allow it to run on severely computationally constrained platforms effectively. This is done using a Markov-chain based control-law transition prediction strategy as well as an intelligent computation cache. Finally, model learning strategies are introduced which can adapt, learn, and compensate unmodeled exogenous disturbances acting on the vehicle online. This technical report also discusses the practical considerations and overall software organization of the predictive control architecture.

author = {Mosam Dabhi and Alexander Spitzer and Nathan Michael},
title = {Aggressive Flight Performance using Robust Experience-driven Predictive Control Strategies: Experimentation and Analysis},
year = {2019},
month = {June},
institution = {Carnegie Mellon University},
address = {Pittsburgh, PA},
number = {CMU-RI-TR-19-08},
keywords = {Aerial Robotics, Robust Nonlinear Model Predictive Control, Belief Propagation, Experience reuse},
} 2019-08-22T14:45:53-04:00