Technology is a key discriminator for Raytheon in bringing the most advanced capabilities to our customers. Raytheon’s research and development of new technologies often occurs in a highly collaborative environment where ideas can originate internally or externally from partners in industry, academia and national laboratories. University partnerships, in particular, involve participation in, and support for, early stage or basic research activities which provides insight into scientific advances — a critical input into the development of technology roadmaps. Raytheon actively partners with leading technologists at more than 75 universities to bring the best minds to bear on developing unique and strategic product solutions for the customer. In this article, we highlight one of these important partnerships with the California Institute of Technology’s (Caltech) Center for Autonomous Systems and Technologies (CAST) programs.

In 2017, Caltech opened CAST to stimulate interdisciplinary research and to accelerate innovation in the way machines help humans achieve scientific, engineering, and humanitarian goals. At CAST, researchers from Caltech and the Jet Propulsion Laboratory are collaborating on the development of autonomous drones, robotics and satellite technology. Raytheon currently sponsors three biologically inspired projects at CAST that align with the company’s core technology interests: Learn to Fly, Swarm Artificial Intelligence and Safe Autonomy.

Learn to Fly 

CAST’s Learn to Fly project is focused on the application of machine learning techniques to address challenging flight control problems and to ensure safe learning. The dynamics of flight can be extremely complex and often difficult to accurately describe using traditional models. For example, a given flight vehicle’s design can make its dynamic motion difficult to characterize with the confidence required to adequately control the vehicle while in flight. Often, an extensive amount of ground testing and characterization are required to develop safe and robust flight control systems for these vehicles. By demonstrating the ability of machine learning algorithms to learn complex aerodynamic behaviors that are not adequately captured with typical modeling approaches, one can fly in more challenging environments where the application of traditional model-based design techniques might not be feasible. Ideally, in the future, the implementation of learning and adaptation of control algorithms will be conducted in real time without the aid of large amounts of simulation data.

Figure 1: Action sequence of a single neural lander performing circles subject to unknown ground effects over a table. The sequence is taken at the first 0, 45, 180, and 235 degrees of the circle.1

The use of learning techniques to modify the control laws of vehicles while in flight can be difficult because the algorithms may not guarantee safe operation of the flight system during the learning process. Learn to Fly research focuses on safe learning. A robust learning algorithm must ensure that the vehicle remains in a safe-state during the learning process for the technique to be truly effective. For example, a system (such as the multirotor vehicle shown in Figure 1) must be prevented from contacting obstacles or the ground while still allowing the learning algorithm to adapt unknown vehicle dynamics. Flight vehicles may also demonstrate rapid departures from stable flight. A safe control strategy must therefore ensure simultaneous learning and stabilization. By fusing concepts from machine learning and control theory, CAST research demonstrates the ability of machine learning techniques to guarantee safety during the learning process. In partnering with CAST, Raytheon is positioned to rapidly mature complex flight systems from development to deployment.

Swarm Artificial Intelligence (AI) 

The second CAST project, Swarm AI, is motivated by collaborative behaviors found in nature that achieve common goals of self-preservation and protection, demonstrating that large groups working together can successfully solve challenges that individuals cannot. The Swarm AI project researches fundamental problems in collaborative operations where individual agents (e.g. drones, robots, satellites, etc.) within a swarm may only communicate with their near neighbors to achieve a common goal. Such configurations are scalable since they do not overload the ability for any one individual to communicate within their network. Another benefit of collaborative behavior is that it can be extremely robust to the loss of individuals if sufficient communication persists among the group. Understanding how to achieve collective consensus, where a large group of communicating agents has the ability to agree on a series of specific tasks or functions, is a prime focus area for existing research applications throughout academia. Taking this research one step further, training swarms of agents to achieve an objective and learning from observed swarm behaviors will help us better understand the benefits and limitations associated with large ensembles of communicating agents.

CAST’s fundamental research in developing artificial intelligence and machine learning techniques for the Swarm AI project is an important step toward enhancing the ability of swarms to solve difficult problems. Example applications of this technology include consensus path planning, distributed decision optimization, and collaborative autonomy. Through a greater understanding of the potential applications of swarm technology, Raytheon will be better able to provide new solutions to service the needs of the future warfighter.

Assured Autonomy 

Figure 2: A heterogeneous set of robots is used to explore, find, and recover a target in the CAST arena. In the foreground is a Rover Robotics Flipper robot capable of traversing rugged terrain. In the background is a modified Segway® Personal Transporter (PT) carrying a quadcopter. The Segway PT is being used as a mobile platform for the longer range, directed exploring quadcopter. Credit: Ames, Ahmadi, and Singletary.

Assured Autonomy, the third CAST project supported by Raytheon, is about forming teams with assured safety objectives.  Similar to Learn-to-fly and Swarm AI, handling unknowns in a safe manner is the key driver of Assured Autonomy. Just as a group of predators coordinate to safely and successfully engage dangerous prey, CAST is addressing ways that teams of heterogeneous robots can work together in an unknown environment to achieve a mission with as little human input as possible, while still providing mission guarantees and robot safety and survivability. Compared to a single robot housing a suite of capabilities (e.g., Curiosity rover on Mars), heterogeneous robots allow for specialization while simultaneously improving robustness and decreasing the cost and complexity of individual robots. These teams of specialized robots enable larger scope missions. An example of this is shown in Figure 2, where a small rugged robot is used in conjunction with a taller robot and flying robot to perform a search mission.

Operating multiple robots with loose mission requirements in an unknown environment is where humans-in-the-loop becomes necessary. Humans have the innate ability to parse measurements and semantics, plan and assign tasks to achieve a mission, while simultaneously considering the safety of the robot. To achieve autonomy from humans, robots explore an unknown environment using perception and mapping. However, exploration comes with risks such as obstacles that could incapacitate a robot. For the robot to remain safe during exploration and mission execution, high level planning and low level control is required, bringing us to the contributions of the work: the abstraction of control barrier functions2 to ensure safety of not only each robot, but also provide task assignments suited for each robot and mathematical guarantees of mission success.

Since a core focus of Raytheon is to both create and demonstrate new products and services for our customers, we are working with our Caltech partners to validate the results of their technological advances through several challenging physical demonstrations. CAST’s Raytheon-sponsored research directly supports the continuing efforts to provide customers with the most innovative solutions available in aerospace and defense. 

— Rob Fuentes
— James Fisher
— Richard B. Choroszucha
— Anthony Marinilli


Segway® is a registered trademark of Segway, Inc.

Shi, Shi, O’Connell, Yu, Azizzadenesheli, Anandkumar, Chung, and Yue: https://youtu.be/FLLsG0S78ik.
A. D. Ames, X. Xu, J. W. Grizzle, and P. Tabuada. Control barrier function based quadratic programs for safety critical systems. IEEE Transactions on Automatic Control, 62(8):3861{3876}, 2017.