Folder: /mnt/datapool/hermes_workspace/projects/follow-me-robot ·
briefed rev 3 — Permobil base + factory R-net (boxed) + 35A brushed controllers + ESP32s
Start a session here (paste into a terminal):
cd /mnt/datapool/hermes_workspace/projects/follow-me-robot && hermesdocs/
base-survey.md · constraints.md · log.md
output/
Follow-Me-Cart-job-description.pdf · base-survey
Hermes: this file is the context for the Follow-Me Robot project assistant.
You are in /mnt/datapool/hermes_workspace/projects/follow-me-robot. Work from here unless told otherwise.
Convert Mike's Permobil power-chair base (M300/M3/M1 — motors, wheels and frame; 24 V batteries and
charger; and the **factory R-net electronics — power module, joystick, cables — kept off the chair and
boxed) into a cargo cart that follows him around the property carrying 100–150 lb** over gravel,
dirt, grass and gentle slopes, hard-capped at 3 mph. Mike also owns **dual brushed 35 A motor
controllers and several ESP32s. Two phases: Phase 1 makes the base drive safely under our own
control (no camera yet); Phase 2 adds camera-based following**. docs/constraints.md is the
authority on this machine — read it, and docs/base-survey.md, before designing anything.
(power module, joystick, cables). The factory-electronics bench test is cancelled — there is nothing
left to test. The cart now has to run on hardware we specify ourselves, not the chair's own.
the 24 V batteries and charger, and any cabling that ran between those parts — and record it in
docs/base-survey.md. Then establish brushed vs brushless from the motor labels or photos: that
one answer decides whether Mike's own dual 35 A brushed controllers + ESP32 can drive this at all
(brushed → our own controllers are viable; brushless → those controllers are unusable and we specify
brushless controllers instead). Specify nothing until the motor type is known.
motor brakes (which hold when power is pulled) and the machine's speed limiting. Both are now
ours to design. Record the brake-release design before anything is built.
Phase 1 done = the six Phase 1 tests in docs/constraints.md §6 pass with the numbers recorded:
each motor drives and steers under our controller, the brakes lock when power is pulled (verified
by pulling it), the hand-held e-stop stops it from full speed in under 1 second **with the software
killed**, it is capped at 3 mph from every input, it carries 150 lb around a 100 ft gravel loop, and it
runs 30 minutes on a charge. Phase 2 done = it follows Mike at 3–6 ft on gravel, stops within 2 ft
when it loses sight of him, and the tracking works in sun and in low light. Every part in any list has
a real part number, a price and a source link.
```
/mnt/datapool/hermes_workspace/projects/follow-me-robot/
.hermes.md this file — auto-loaded project rules
README.md human quickstart (how to start THIS assistant)
src/ source code for the project
scripts/ runnable one-off scripts + generated tools
schemas/ JSON schemas for every custom tool (<name>.schema.json)
output/ generated artifacts, reports, exports
docs/ notes, decisions, references
```
(A per-project venv/ cannot exist on this share — build venvs at /home/mike/.venvs/<name>.)
Shared, project-independent facts (host, storage, runtime) live in
/mnt/datapool/hermes_workspace/ENVIRONMENT.md — read it instead of assuming.
1. Save every generated script, schema, and output under /mnt/datapool/hermes_workspace/projects/follow-me-robot/ —
never leave artifacts in ~ or the scratch dir.
2. Every custom tool ships as two files in the same change:
scripts/<tool_name>.py and schemas/<tool_name>.schema.json.
Copy the working pair at /mnt/datapool/hermes_workspace/templates/example_tool.py +
example_tool.schema.json as the starting shape: pure function, stdin JSON →
stdout JSON, no side effects on import.
3. Test before marking anything complete. No Docker here. **Build venvs on LOCAL disk, never in the
project** — /mnt/datapool is CIFS and refuses the lib64 symlink that venv creates, so
python3 -m venv venv inside the project dies with Errno 95: Operation not supported, and
--copies does not avoid it. It leaves a venv/ holding lib and no bin/python, and anything
depending on it then cannot run. Use
/usr/bin/python3 -m venv /home/mike/.venvs/<name> and call that interpreter by absolute path
(add pypdfium2 too if the tool rasterises images). Otherwise run the script and show real
output. Paste actual command output, never a description of it.
4. Keep terminal explanations to a single copy-and-pasteable command per step.
5. Every part in the list needs a real part number, a price and a source link — no placeholder parts.
That applies to used and salvage parts too: a named listing, a real asking price and a link,
or it does not go in the list.
6. Phase discipline. Phase 1 has a ceiling of about $400 and a 50 lb payload target. Do not
design Phase 1 to carry 150 lb — an over-specified first build costs Phase 2's budget and proves
nothing sooner. Any part over about $40 must carry a named cheaper alternative in the same table.
7. Safety is a requirement, not an option. The list in docs/constraints.md §5 is mandatory:
a hand-held e-stop that cuts power independently of software, failsafe stop on lost sight or
signal, a hard speed cap, no autonomous mode, nothing above 48 V DC. Any design that omits one
must call that omission out with a reason — never quietly leave it out.
8. Safety-critical and structural choices go to Mike as a written decision, with the numbers and
the alternatives, and clearly flagged. Do not settle a brake, e-stop or tipping-limit question
inside a parts list.
9. Design only. Nothing is ordered, bought, wired or paid for until Mike says so. Salvage parts
may be recommended and linked; they are not to be purchased.
10. Keep the base's brakes fail-safe. The drive motors carry normally-closed electromagnetic
brakes that lock when unpowered — that behaviour is the machine's primary safety system. Never
design a circuit that holds the brakes open when power is lost, never bypass them, and never test
the machine with the free-wheel levers released. Any change to brake wiring is a written decision
for Mike (rule 8), not an implementation detail.
11. Wheels off the ground for every first power-up. A 300 lb-rated chassis with 120 A of motor
behind it does not care what is in front of it.
Hermes + delegate_task subagents; web search for parts and pricing; Python/OpenCV for person detection; project-map diagrams for the architecture.
docs/constraints.md, docs/parts-list.md, an architecture diagram, tracking code under src/
This block is injected into every project's .hermes.md by bin/new-project, so
host facts live in ONE place. Update here, not per project.
mike//10.0.0.5/datapool mounted at /mnt/datapool (cifs, vers=3.1.1, uid/gid 1000, rw). There is NO /mnt/t130_storage on this host.
/mnt/datapool/hermes_workspace//usr/bin/python3 (3.14.4, venv + pip both work). ⚠ Inside a Hermes session, bare python3 resolves to Hermes' own bundled runtime
(3.14.7) which has no pip — python3 -m pip install x fails with
"No module named pip". Always build project venvs explicitly:
/usr/bin/python3 -m venv venv && ./venv/bin/pip install ...
huggingface_hub 2.4.0, mcp 2.3.0, numpy 2.5.4 (cp314 wheel) and tokenizers 0.23.3 all resolve — the HF/MCP stack is fine
on this Python, no downgrade needed.
~/.hermes/ (config, skills, cron, sessions, state.db)main). Two local configs are mandatory on the SMB share, not cosmetic: core.filemode=false (the share forces
0755) and core.trustctime=false — CIFS ctimes are unreliable, and with them
trusted git add can treat a just-written file as unchanged and commit a **stale
blob**. Commit when a unit of work is finished:
cd /mnt/datapool/hermes_workspace && git add -A && git commit -m "...", then
confirm git status --short is empty — if it still lists files, re-add and commit
again before trusting the commit.
9616162615e7 mirrors the workspace to gdrive:3050-HOME/hermes_workspace (one-way up, .git included, so the
Drive copy is a real clonable repo). Script ~/.hermes/scripts/backup_hermes_workspace.sh,
log ~/.hermes/cron/output/hermes-workspace-backup.log. It prints nothing on
success and pings Telegram only on failure.
rclone sync against a missing mount would delete the Drive copy, so it gates
on the mount, a sentinel file, and a minimum file count, and caps deletions.
3050-to-Drive bisync job (sync_to_drive.sh, which covers $HOME/{book,hermes-3050,agent-os,projects,bookapp}).
Two sync jobs touching one tree is how you get conflicts.
Verified against live command output 2026-10-30. Re-verify with:
hostname; ip -4 addr; mount | grep datapool; python3 -V; node -v; docker --version
When work stops: update Status (state + next step) in this file, then log one
line in /mnt/datapool/hermes_workspace/projects/follow-me-robot/docs/log.md with date + what changed + what's next.
Built live from your own files — nothing here is typed by hand.