Capabilities

Assurance across the autonomy stack.

Four connected capabilities, one question: can the mission trust this autonomous system when the environment is actively working against it?

Assured Autonomy

Multi-agent systemsSwarmsGround & air autonomy

Mission problem: Autonomous and multi-agent systems must keep the mission going when sensors, agents or links are compromised or degraded.
Our approach: Reinforcement learning and adversarial methods for attack-resilient planning and control. Formal verification of mission-critical decision logic. Runtime assurance that detects unsafe behavior and hands control to a verified fallback.
What we build: Resilience analyses, verified decision and control components, and runtime monitors integrated with ROS 2 and PX4 autonomy stacks.

AI T&EAdversarial robustnessSWaP-limited compute

Mission problem: AI on small platforms faces adversarial inputs, distribution shift and tight size, weight and power budgets, often with no link back to a human.
Our approach: Adversarial robustness testing and hardening. Confidence-calibrated outputs so operators know when to trust a call. Energy-aware machine learning that fits onboard compute without giving up accuracy.
What we build: AI test and evaluation harnesses, robustness assessments, and hardened models ready for edge hardware.

5G / NextGDDILContested spectrum

Mission problem: Tactical and 5G-based networks are jammed, spoofed and misconfigured. When the link fails, the autonomy that depends on it fails too.
Our approach: AI-driven anomaly and intrusion detection for wireless links. Formal policy assurance for programmable networks. Causal reasoning that finds root causes when the network degrades.
What we build: Network assurance tools, testbed evaluations on real radios, and communication strategies built for denied and intermittent conditions.

Group 1–3 UASPX4 / MAVLinkSupply chain

Mission problem: Small UAS built on open-source autopilots inherit the software supply chain and protocol weaknesses of those stacks.
Our approach: Vulnerability analysis and threat modeling of open-source flight stacks. Formal verification of MAVLink and autopilot logic. Onboard sensor-fusion monitors that flag spoofing, faults and damage while the aircraft is flying.
What we build: Security assessments, verified flight software components, and monitoring software validated in flight on our own platform.

Emerging area

Quantum computing security. Our founders also lead research on the security and verification of quantum and hybrid quantum-classical systems.