GlasgowAstro.Explorer AI Agent 🌌👽

Welcome to GlasgowAstro.Explorer, an intelligent explorer designed to traverse alien biomes and collect virtual samples for scientific analysis. This project leverages the smolagents framework to create an agent that can move, investigate, and catalog extraterrestrial life forms and ecosystems - all from the safety of your computer.

Current Challenges & Observations

• The agent sometimes loops endlessly or revisits the same planet positions instead of covering new ground.
• Certain models (e.g., deepseek-r1:7b, qwen2.5-coder:7b) struggle to follow tool-calling rules, skipping or misusing available tools.
• The agent occasionally invents discoveries or “simulates” exploration rather than relying on actual tool output.
• Large models (e.g., qwen3-coder:30b) perform better reasoning but can be too slow or memory-intensive on local hardware.

Example run

Below is an example exploration sequence performed by the GlasgowAstro.Explorer agent.

First area scan:

Second area scan:

First discovery:

Second discovery

Discovery summary

Technologies

• Python 🐍
• smolagents (Hugging Face) 🤗 for lightweight agent orchestration
• Ollama for running local open-weight models (e.g., Qwen, DeepSeek etc.).
• Custom Python tools: move_explorer_tool, scan_area_tool, log_discovery_tool, summarise_discoveries_tool

Alien life

Otteroo