DARPA Learning Applied to Ground Robots (LAGR) Project (Concluded)

The National Institute of Standards and Technology (NIST) Intelligent Systems Division (ISD) participated in the Defense Advanced Research Project Agency (DARPA) Learning Applied to Ground Robots (LAGR) Program for a three year periods. During the first 18-month project phase, NIST developed learning algorithms and became familiar with the LAGR vehicles and DARPA test methods as a standard participant in the program, albeit outside the competition.  During the second 18-month project phase, DARPA suggested that all LAGR teams participate in a “best-of” effort by supplying NIST with their best performing, most complete, and most promising perception and/or control algorithms.

The National Institute of Standards and Technology (NIST) Intelligent Systems Division (ISD) participated in the Defense Advanced Research Project Agency (DARPA) Learning Applied to Ground Robots (LAGR) Program for a three year periods.  During the first 18-month project phase, NIST developed learning algorithms and became familiar with the LAGR vehicles and DARPA test methods as a standard participant in the program, albeit outside the competition.  During the second 18-month project phase, DARPA suggested that all LAGR teams participate in a “best-of” effort by supplying NIST with their best performing, most complete, and most promising perception and/or control algorithms.  NIST developed an application programming interface (API) and a control framework to support “plug-and-play” of all algorithms into a vehicle controller for rapid integration and testing to support DARPA LAGR evaluations. In total, there were 27 tests, some with two parts. NIST participated in 21 tests.  These included: Phase 1 tests 1-13 and Phase 2 test 14 which ran a pure Phase 1 system. NIST did not participate in tests 15 – 18 and 23. NIST provided “best of” vehicle controllers for tests 19-22 and 24-27.  During the time in which the tests in which NIST did not participate were run, ISD developed a DARPA-requested standard color scheme for operator control units so that all teams’ controller output provided identically mapped costs to simplify team-to-team comparison. More importantly, during the non-participatory period, NIST developed the “plug-and-play” framework known as the Best-of LAGR (BLAGR) controller.  The outcome of this effort allowed teams to integrate algorithms that abided by the NIST-developed API to be integrated into the BLAGR with relative ease.  NIST then participated in several tests with versions of the BLAGR controller, comparing results to other teams’ controllers.