You are here

Precision Control of High-speed Autonomous Vehicles under High Disturbances


OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy, Integrated Network System of Systems OBJECTIVE: This project seeks the development and demonstration of algorithms that support near-optimal control of autonomous high speed aerial vehicles in real time, with precision, and in challenging and adversarial environments. DESCRIPTION: Unmanned Aerial Systems (UAS) used by the Army may be subject to harsh conditions in hostile environments. They need to be able to sense heavy disturbances in their environment that affect their operations, instantaneously adjust to overcome their impact. Furthermore, they should form and track a mission supporting trajectory in real time with speed and agility. Control systems that directly integrate feedback from complex inertial sensors, such as high-end inertial measurement units, or novel odometry systems, and semantic feedback from exteroceptive sensors, such as cameras have the potential to substantially increase the maneuvering capability of high-speed vehicles used or envisioned by the Army. Such measurements can be used in the feedback loop to instantaneously adjust controls to overcome disturbances, as well as predicting abrupt changes in the disturbances and issue predictive control mechanisms. This approach could enable safe and effective operation of for small UASs under excessive wind, abrupt changes in atmospheric pressure due to effects such as blast waves, and occurrence of obstacles which may not be known in advance. This project seeks the development and demonstration of algorithms that can control autonomous aerial vehicles with precision in challenging and adversarial environments listed above, using integrated real-time information from precision inertial sensors and high-frame-rate cameras. The trajectories are pre-determined by the mission in terms of a sequence of waypoints, but they can be subject to small changes based on real-time information acquired by sensors. In general, models describing such systems are complex, and real-time generation of time optimal control policies is challenging. Incorporation of data driven approaches using innovative machine learning algorithms could provide acceptable near-optimal solutions. The research also involves integration of information from a variety of sensors into a from that can be used by the controllers to ensure the stated goals. The proposed algorithms should be implementable over computing hardware that can fit the platform of choice for the demonstration. Sensors that can provide the required information should be specified and the impact of possible gaps in their commercial availability should be identified. It is expected that the computed control laws can provide a performance within 20% of is preselected values of the trajectories and the instantaneous velocities as obtained from simulations and analysis. The demonstration should produce tracking errors no larger than 20 centimeters over the planned trajectory with wind conditions less than 20 miles/hour and have the errors stay bounded under more challenging conditions. A UAS of four vehicles should be able to perform agile movements as required by the control law at a speed more than 15 miles/hour, while reaching the maximum speed of the platform in favorable parts of the trajectory. PHASE I: During Phase I effort, the proposed control algorithms will be completely specified and validated using simulations over realistic scenarios. The theoretical underpinnings of the proposed algorithms should be discussed with technical rigor, accompanied with their analysis of stability, safety and convergence conditions. PHASE II: Four or more or small-scale prototype vehicles with sensor, computation and control units will be designed based on the numerical model and design methodology developed in Phase I, technologies. The prototype devices can be built on commercially available state of the art small rotary wing quadcopters. If applicable, performers are encouraged their own designed crafts with comparable or better performance than commercially available units. Technical risks will be identified and plans for minimizing these risks will be devised. PHASE III DUAL USE APPLICATIONS: Phase III effort will explore opportunities for integrating developed technologies into various UAS and weapon systems used by the Army. REFERENCES: 1. W. Sun, G. Tang, and K. Hauser. Fast UAV trajectory optimization using bilevel optimization with analytical gradients. In 2020 American Control Conference (ACC), pages 82–87. 2. D. Mellinger and V. Kumar. Minimum snap trajectory generation and control for quadrotors. In 2011 IEEE International Conference on Robotics and Automation, pages 2520–2525 3. S.Meyn. Control Systems and Reinforcement Learning, Cambridge University Press, 2022. KEYWORDS: UAS, Trajectory Planning, Time Optimal Control, Reinforcement Learning
US Flag An Official Website of the United States Government