Belleau Labs
Back to contracts

Tactical Funding Increase (TACFI): Offline Learning and Counter Artificial Intelligence for Autonomous Aircraft Combat Operations

AFWERX · AFWERX TACFI · AFWERX

AI-Readiness Score
20/25
Pathway Speed
5/5
Timeline Realism
3/5
Problem Framing
4/5
AI / ML Fit
5/5
Award + Transition
3/5

Award

$1,900,000
Award ceiling
TOYON RESEARCH CORPORATION
Awardee
Posted September 12, 2024

Description

Reinforcement learning (RL) consistently produces controllers that exceed human performance on complicated tasks in control and strategy. Despite this promise, their widespread use is limited by a few important problems. First, they are incredibly compute intensive. Popular RL algorithms do not allow the reuse of data during their learning. This problem results in inflexible controllers because controllers cannot be easily modified for a new task, and this constraint limits the number of controllers that can be deployed because each controller takes immense resources to create. Last, this inflexibility results in suboptimal controllers because controllers cannot easily learn from different sources of data, which includes examples from expert human operators. The second issue is that most controllers learned by RL are susceptible to counter AI attacks, which can force a controller to fail catastrophically. Toyon Research Corporation has developed new training methodologies that enable data reuse and the ability of RL algorithms to learn from related tasks and Counter AI attacks. On this effort we will extend our work to support air-to-air combat operations and live flight tests.

Score Rationale

TACFI is one of the top-tier fast-track instruments (highest score), and the problem is genuinely AI-core — offline RL, data reuse, adversarial robustness, and autonomous air combat are all deeply AI-shaped with a named incumbent (Toyon) and live flight test endpoint. The award ceiling of $1.9M lands in the mid-tier range with an implied Phase III transition pathway inherent to TACFI structure but not explicitly named here, and the unknown response deadline prevents a higher timeline realism score despite the prototype scope being well-bounded. Problem framing is strong given the specific RL failure modes called out, but loses one point because end-user operational integration details and concrete success metrics are not spelled out in the excerpt.

Source

View original posting
Back to all contracts