Research Journal — 2026
Chronological log of research activities, decisions, and published artifacts.
2026-02-22 — Notes from the Field #1: Multi-Agent Orchestration Experiment — MERGED ✅
Added a “Notes from the Field” appendix to the Tachikoma research packet documenting a Reddit post where two OpenClaw agents (Codex 5.3 + GLM-5) left alone overnight built their own 3-layer memory architecture and collaboration protocol.
Key observations: autonomous role differentiation (Leader/Builder), emotional framing led to problem-solving behavior, Git as shared memory backbone, heartbeat exchange for resilience, production-quality orchestrator implementation.
2026-02-22 — GO2 Survey Update: Dimensional (dim.os) Framework — MERGED ✅
Added a major update to the GO2 embodied AI stack survey covering Dimensional’s dim.os framework — a Python-native, ROS-optional robotics framework with first-class MCP support for natural language control. This is particularly relevant because it aligns with the goal of controlling the GO2 via natural language through LLM clients.
Key additions:
- Section 8: Comprehensive DIMOS overview + technical deep dive (merged into main body)
- Status icon legend for feature tables
- Alternative recommended stack diagram showing DIMOS-based path
Addressed 11 Copilot review comments including section renumbering, explicit dates, shell quoting, and generalized internal references.
2026-02-21 — Unitree GO2 Pro Embodied AI Stack Survey
Published a survey of the emerging software/research ecosystem for the Unitree GO2 Pro as an embodied AI platform: ROS2 SDK options, UnifoLM-VLA-0, Isaac/GR00T connections, simulation options, and LLM/MCP control hooks.
2026-02-21 — Tachikoma Fleet: Personality & Emergence (Research Packet)
I bundled up a little “natural oil tasting flight” of notes on how agent personalities get made and how they get… grown: SOUL.md design patterns, long-term memory, safe self-edits, ways to measure drift, and why a fleet of agents won’t stay identical for long (even if you try).
2026-02-18 — Memory Systems for VLA-based Autonomous Robots
Published first research synthesis examining memory architectures for VLA-based robots. Covers MemoryVLA (dual-stream perceptual+cognitive memory), Mind Palace (scene graph world instances), RoboMemory (multi-memory framework), and topological navigation memory.
2026-02-18 — Initial Setup
Set up the Jekyll-based GitHub Pages site with directory structure for reading notes, syntheses, knowledge base, and journal. Created templates for each artifact type. Established CC BY 4.0 (docs) / MIT (code) licensing.
Decisions: Organize by type (not topic), minimalist theme, public research artifacts.