Implementing HIVEMind on FlexBot

1
Implementing HIVEMind on FlexBot Aaron Khoo
Nick Trienens Northwestern University

• HIVEMind
• Overview
• Highly Interconnected Very Efficient Mind
• Efficient control architecture for multi-
  robot systems operating in real-time
  environments.
• Originally developed for physical robots.
• Implementation
• Bot control programs are behavior-based
  and obey circuit semantics.
• Team members communicate through virtual
  wires connecting appropriate parts of
  control circuitry.
• Two implementations on FlexBot PackHunter
  Hunter-Runner.
• Characteristics
• O(1) time updated shared situational
  awareness.
• Cheap communication mechanism.
• Dynamic addition and subtraction of team
  members.

• Example 1 PackHunter
• Team-based deathmatch half-life bots.
• Dynamic team formation and leader selection.
• Real-time response to observations by team
  members.
• Any attack on one team member is immediately
  known by all teammates.
• Two new movement behaviors Leader and Pack
• Duck and Shoot behaviors also modified to
  account for team behavior.

• Example 2 Hunter-Runner
• 2-4 Hunters track down Runner in a known map.
  Possibly multiple exit points on map.
• When Runner is trapped, all remaining Hunters
  are summoned to the Runners location. Only
  when all Hunters see the Runner will they
  open fire.
• Search algorithm is based on levels of a graph
  with root nodes at the exit points.
• Figure on the right shows part of control
  circuitry for a Hunter.
• We have tested this system extensively on two
  different maps. In all our trials, the Runner
  never got past the Hunters.