DARPA – Transfer from Imprecise and Abstract Models to Autonomous Technologies (TIAMAT) – DARPA-PS-23-13

DARPA is seeking innovative proposals for novel techniques to rapidly transfer autonomy from simulation to the real world such that it is

1.  robust to the quick and inevitable changes in dynamic environments and
2. adaptable to a variety of platforms and domains.

TIAMAT is structured as a two-phase program

• Phase 1 – Developing sim-to-sim autonomy transfer methods – 18 months
• Phase 2 – Developing sim-to-real autonomy transfer methods – 18 months

Technical Challenges

• how to conduct autonomy transfer under a restricted time budget such that it is robust to sim-to-sim and sim-to-real gaps in agent observations, actions, transitions, and goals
• how to refine the models and simulations used based on agent experience.

To elaborate on these challenges, consider the framing of autonomy transfer and abstraction refinement in the context of decision processes.  This framing is for illustrative purposes and proposers are free to use other formalisms and forms of autonomy, including domains from computer vision and related areas where decision processes are not commonly employed.

Read more here and here.

Document

TIAMAT Funding Announcement: DARPA-PS-23-13