notesum.ai

Published at December 6

TeamCraft: A Benchmark for Multi-Modal Multi-Agent Systems in Minecraft

cs.AI

cs.CL

cs.CV

cs.MA

Released Date: December 6, 2024

Authors: Qian Long¹, Zhi Li¹, Ran Gong¹, Ying Nian Wu¹, Demetri Terzopoulos¹, Xiaofeng Gao²

Aff.: ¹University of California, Los Angeles; ²Amazon AGI

Arxiv: http://arxiv.org/pdf/2412.05255v1

Refer to caption

	Building	Clearing	Farming	Smelting
# Action Sequences	2 – 6	2 – 9	2 – 7	2 – 8
# Agents	2 – 3	2 – 3	2 – 3	2 – 3
# Tools	–	1 – 4	–	1 – 4
# Scenes	6	5	4	5
# Base Types	10	11	9	11
# Furnaces	–	–	–	1 – 2
# Target Block Types	19	16	3	13
# Target Block Counts	5 – 12	4 – 9	2 – 14	1 – 4
# Fuel Types	–	–	–	12
# Resource Types	–	–	–	20
# Dimensional Shapes	2	2	2	1
# Placement Shapes	7715	12724	13188	8885
# Total Demonstrations	14998	14641	14815	10803
# Test Set	50	50	50	50
# Generalization Set	200	200	150	200
# Generalization Conditions	4	4	3	4