WISE: A Long-Horizon Agent in Minecraft with Why-Which Reasoning

2024-05-13 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

WISE (Which-Why Informed Semantic Explorer), a long-horizon agent framework for Minecraft, addresses limitations of existing LLM-augmented hierarchical agents, specifically low-level controller bottlenecks due to repeated execution failures and decoupled memory/reasoning. It introduces a Causal Event Graph to augment episodic memory with explicit causal structures, linking observations to task relevance for robust recall under viewpoint changes and opportunistic task reordering. WISE also features an Opportunistic Task Scheduler for dynamic subtask reprioritization and a multi-scale progressive exploration strategy. Experiments demonstrate WISE's effectiveness, showing a 30% increase in sequential sparse task success with 26.4% lower completion time, and a 44% increase in adaptive non-sequential task success with 42.5% less completion time compared to MrSteve. It further achieved 97% map coverage on a 128x128 real-world map, significantly outperforming MrSteve's 67%.

Key takeaway

For AI Scientists developing embodied agents for complex, open-ended environments like Minecraft, traditional hierarchical LLM-augmented agents often bottleneck at low-level control due to rigid memory and planning. You should integrate semantic causal memory and opportunistic scheduling to enable adaptive decision-making and robust recall. Consider multi-scale exploration to ensure comprehensive observations, significantly improving task success and efficiency in sparse, long-horizon scenarios.

Key insights

WISE unifies exploration, memory, and planning in Minecraft agents via causal reasoning for adaptive, efficient long-horizon task execution.

Principles

• Causal reasoning enhances episodic memory beyond visual similarity.
• Dynamic task scheduling improves efficiency by exploiting opportunities.
• Multi-scale exploration ensures comprehensive environmental coverage.

Method

WISE uses a Causal Event Graph for semantic memory, an Opportunistic Task Scheduler for dynamic subtask prioritization, and a multi-scale progressive exploration strategy (quadtree, frontier, Voronoi) for comprehensive observation acquisition.

In practice

• Implement a Causal Event Graph for semantic memory in embodied agents.
• Integrate an opportunistic scheduler for dynamic task re-prioritization.
• Employ multi-scale exploration for efficient environment coverage.

Topics

• Embodied AI Agents
• Minecraft Environment
• Causal Reasoning
• Large Language Models
• Episodic Memory
• Task Scheduling
• Multi-scale Exploration

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• WISE: A Long-Horizon Agent in Minecraft with Why-Which Reasoning — WISE enhances long-horizon agents with a Causal Event Graph and Opportunistic Task Scheduler for adaptive decision-making in complex environments like Minecraft.
• Learning to Choose: An Empowerment-Guided Multi-Agent System with semantic communication for Adaptive Method Selection — Reliable autonomous learning in multi-agent systems requires preserving action-outcome fidelity through semantic checkpoints to prevent drift.
• OCR-Memory: Optical Context Retrieval for Long-Horizon Agent Memory — Visual encoding of agent trajectories enables long-horizon memory with reduced token costs and hallucination.
• The cognitive companion: a lightweight parallel monitoring architecture for detecting and recovering from reasoning degradation in LLM agents — A parallel monitoring architecture can detect and mitigate LLM agent reasoning degradation with minimal or zero overhead.
• LongMINT: Evaluating Memory under Multi-Target Interference in Long-Horizon Agent Systems — Current memory-augmented agents struggle with interference and multi-target reasoning in long, dynamic contexts.
• Remember Me, Refine Me: A Dynamic Procedural Memory Framework for Experience-Driven Agent Evolution — ReMe dynamically manages LLM agent procedural memory, enabling smaller models to surpass larger, memoryless counterparts through experience-driven evolution.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.