MATE-BT: A Behavior-Tree-Based Framework for Hierarchically Decentralized Mission Autonomy in Heterogeneous Multi-Robot Teams
Authors: Sunbeen Lee, Hoseon Gi, Jik Hur, Inmo Jang
Organizations: Department of Artificial Intelligence, Korea Aerospace University, Goyang, Republic of Korea · Korea Aerospace Industries, Ltd., Sacheon, Republic of Korea
Multi-robot teams executing predefined missions must adapt to runtime changes and sustain mission management despite robot failures. This paper proposes MATE-BT (Mission, Allocation, and Task Execution with Behavior Trees), a hierarchically decentralized framework that separates mission execution, allocation, and task execution. An active onboard leader manages mission progression and team-level task assignment, with authority transferable to another leader-capable robot after failure. Interchangeable decentralized methods resolve target assignments, while each robot adapts its local behavior through planning and acting with behavior trees. Webots simulations and real-robot experiments demonstrate mission continuity despite new targets and robot failures, including active-leader failure. In the physical experiments, all MATE-BT computation runs onboard. Hardware-in-the-loop resource measurements further support onboard execution on Jetson Orin Nano and Raspberry Pi 5. By updating only the affected layers without full-plan regeneration, MATE-BT maintained a response time of approximately 130 ms with 30 robots and 120 targets. Comparisons of four target-assignment plugins reveal scenario-dependent trade-offs between mission completion time and total travel distance.
Figures & tables
Fig. 1: Overall architecture of MATE-BT. The team-level Swarm BT runs on leader-capable robots, while the robot-level Agent BT runs on every robot. The Agent BT invokes robot-specific action primitives for task execution. Shaded blocks mark allocation, which spans both BTs: task assignment in the Swarm BT and target assignment in the Agent BT. Dashed blocks are supplied by the user for a given mission and robotic platforms. Blue lines are ROS interfaces between robots, and black arrows are blackboard reads and writes within a robot.
Fig. 2: Representative snapshots and mission timelines of the fire-suppression-and-rescue mission in Webots. Panel (a) shows the environment and the robot team, panels (b)–(e) show the normal mission sequence, and panels (f) and (g) show continued execution after a Fire UGV failure and a Rescue UGV failure, respectively. Panels (h) and (i) show the corresponding mission timelines for the normal and breakdown scenarios. Vertical dashed lines mark phase transitions, colored bars indicate robot tasks, the upper bands show the number of active fires and the mission-managing robot, and bottom markers denote victim events. In (i), crosses mark robot breakdowns and hatched regions indicate subsequent robot unavailability.
Robot type
Quantity
Priority
Executable tasks
Rescue UGV
1
0
rescue, return
Fire UGV
3
1–3
suppress, escort, rescue, return
UAV
2
–
search, return
TABLE I: Heterogeneous Robot Team Configuration
Fig. 3: Breakdown scenario on four UGV robots, seen from the overhead camera with augmented overlays. Red discs are fires, yellow squares are victims, a lost robot carries a cross, and the base is at the lower left. (a) The team starts the Rescue phase with Rescue UGV 1 escorted. (b) Fire UGV 1 is lost. (c) The leader, Rescue UGV 1, is lost. (d) A Fire UGV rescues the last victim. (e) The Suppress phase clears the remaining fires. (f) The Return phase ends with the surviving robots at base. (g) The mission timeline for this run, with the conventions of Fig. 2 .
Metric
Normal
Breakdown 1
Breakdown 2
Runs completed
10/10
10/10
10/10
Completion time (s)
252.74
± 12.28
323.61
± 42.32
366.92
± 33.63
Handover latency (s)
–
2.44
± 0.08
2.46
± 0.06
Task reassignments
0.0
± 0.0
4.0
± 0.0
1.0
± 0.0
Target reassignments
63.4
± 6.5
77.6
± 15.6
85.6
± 10.5
Task BT expansions
37.4
± 0.5
51.2
± 1.6
40.0
± 0.7
TABLE II: Comparison of Failure Responses Across Scenarios
Raspberry Pi 5
Jetson Orin Nano
Role
CPU (%)
RAM (MiB)
CPU (%)
RAM (MiB)
Idle (ref.)
6.8
1419
5.4
1851
Active leader (Swarm BT + Agent BT)
29.3
1586
20.7
2006
(+22.5)
(+2.1)
(+15.3)
(+2.0)
Leader-capable (Swarm BT + Agent BT)
28.4
1534
20.2
1952
(+21.6)
(+1.5)
(+14.8)
(+1.3)
TABLE III: Onboard Resource Utilization per Role
Figure 7
Fig. 6: Allocation cases with spread fires in Case 1, clustered fires in Case 2, sequential fire batches in Case 3, and large fires requiring cooperation in Case 4.
New York University, New York, USA · U.S. Army Combat Capabilities Development Command, Army Research Laboratory, Adelphi, MD 20783, USA · Vanderbilt University, Tennessee, USA +1