Our research > Cybersecurity research > Design and control of cyber physical systems

​research and development in Design and Control of Cyber Physical Systems

Peraton Labs has an active research and development program in the design of cyber physical systems (CPS) and their controllers. Leveraging deep expertise in nonlinear optimization, robotics, AI, optimal control, and mathematical logic, we develop novel tools, methods, and solutions for improving the design and control of robots, uncrewed vehicles, and autonomous systems. Sponsored under past and current DARPA programs, our work has resulted in high-profile research publications as well as transition-ready prototypes. 

​Adaptive Control with AI (ACAI): Our AI control technology, called Adaptive Control with AI (ACAI), allows robots to detect unexpected conditions, learn new dynamics, and adapt controllers to maintain safe operation and make progress towards mission goals—even under conditions of high uncertainty. 

• Supported under the DARPA Learning Introspective Control (LINC) program, ACAI assures safe operation of diverse robotic systems in challenging environments and unforeseen conditions, while also providing guidance and situational awareness to human operators.
• ACAI has been successfully tested on a damaged, robotic Army ground vehicle operating in difficult terrain and on a crane performing simulated ship-to-ship replenishment in heavy seas; see our press release on Peraton Labs Demonstrates AI Control Capability for Safe Robot Operations.
• ACAI combines innovative methods to deliver a minimal, efficient, and mathematically-grounded system that consistently exceeds safety benchmarks, surviving destabilizations for times greater than expected. ACAI shifts the burden of resilience from human operators to the robotic system.
• Learn more about ACAI in our presentation on "Adaptive Control with AI: An Experience Report from DARPA LINC Program," Imperial College, London, September 2025, including details on these innovative methods:​​
  • ACAI safety controller guarantees stability against worst-case disturbances using ISAACS reachability methods
  • ACAI functionality controller preserves operator intent with Gaussian Process learning
  • ACAI repeated LQR approach stabilizes crane payloads on moving platforms

Automating Design and Control: We have developed automated, intelligent methods and toolkits for efficient, accurate, and secure design and control of CPS and autonomous systems. Our AI-driven approach significantly shortens the solution cycle, greatly reduces effort and cost, guarantees operational accuracy and safety, and enhances solution innovation.

• Funded under DARPA’s Symbiotic Design of Cyber Physical Systems (SDCPS) program, our solutions ensure “correct-by-synthesis” designs and also reduce the time from concept to deployment of CPS from years to months. Our methods not only ensure that operational environment and mission requirements are met, but can also quickly find solutions that provide design robustness to changes in the requirements as well as solutions for redesign and repair to accommodate degradations in the system or alterations in the requirements.
• AI-Mediated Exploration of Design (AIMED): Our AIMED technique is extremely efficient at learning the promising areas of the CPS design space and rapidly identifying candidate solutions. Design spaces are high-dimensional cross-products of discrete and continuous spaces representing numerous design variables. Human designer’s intent may not be concretely articulated and it can take minutes to hours to evaluate the performance of a candidate design.
  • ​AIMED automatically discovered high-scoring, novel designs for unmanned airborne vehicles (UAVs), including designs unencumbered by biases of planarity and symmetry such as UAVs with non-coplanar propellers and asymmetric wings. AIMED was also successfully applied to the design of unmanned underwater vehicles (UUVs). 
  • Read more about AIMED in AI-Mediated Exploration of Design: An Experience Report, presented at CPS-IoT Week Workshops ’23, including details on the following key innovations:
    • Deformable connectors which eliminate an important type of discreteness from design spaces
    • Inverse specification  method based on inverse reinforcement learning which infers human intent by asking a small number of simple preference questions
    • Gaussian mixture models which allow completion and repair of designs and find a diversity of solutions
• Constrained Optimization with Neural Networks, Mixed Integer Linear Programming, and Active Learning (CNMA) is the patented, sample-efficient method that forms the core of our automated approach. CNMA leverages a unique combination of machine learning, neural network models, and efficient and parallelized algorithms to produce fully compliant solutions substantially faster than competing approaches and to find solutions when other approaches fail.
  • ​CNMA performed significantly better than conventional optimization methods in evaluations on seven nonlinear design problems. CNMA produced solutions for all problems, including many which conventional methods could not solve. When conventional methods produced solutions, CNMA improved upon their performance by up to 87%.
  • Read more about CNMA in Fast Design Space Exploration of Nonlinear Systems: Part I, published in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, including information on:
    • CNMA's powerful solution for the inverse design problem which, given requirements or constraints on an output, finds an input that meets the requirements and also optimizes an objective function
    • CNMA's efficiency in leveraging neural networks, mixed integer linear programs, and a learning-from-failure feedback loop to sample only that part of the design space relevant to solving the inverse problem
    • CNMA's parallel version which improves the efficiency and quality of solutions over the sequential version and steers away from local optima​