# Follow-Me Robot — log

- 2026-10-31 scaffolded from templates/PROJECT.*.tmpl
2026-11-01 — Briefed. Mike answered four scoping questions: job = haul tools and gear around the
property while following him; terrain = gravel, dirt, grass, gentle slope; payload = 100-150 lb;
budget = a few hundred dollars to prove it, then scale. Wrote docs/constraints.md (the authority):
four answers as design consequences, the budget-vs-payload collision and the used-mobility-scooter
salvage route, a two-phase split with six Phase 1 acceptance tests, mandatory safety requirements,
and the follow-sensor options to compare. Updated .hermes.md Objective, Status, "done" and rules
6-9 (phase discipline, safety, safety choices go to Mike, design only), and corrected the venv rule.
Next: docs/follow-sensor-tradeoff.md, then the Phase 1 parts list against a $400 ceiling.

2026-11-01 (later) — RE-BRIEFED, rev 2. Mike revealed the base already exists: a Permobil power
chair (M300/M3/M1) with motors, wheels, frame, and BOTH batteries and charger — but with all the
electronics removed. Rewrote docs/constraints.md against the real base: the drivetrain problem is
solved (300 lb-rated outdoor chassis, mid-wheel drive, fail-safe electromagnetic brakes), the R-net
CAN route is dead (needs the power module, which is gone), and the new hard fork is brushed vs
brushless motors — documented Sabertooth 2x32 path with P1/P2 brake outputs if brushed, different
plan if brushless. Phases re-cut: Phase 1 = make the base drive safely under our control (no camera
yet), Phase 2 = add following. Added docs/base-survey.md as the fill-in form for the physical facts.
Updated .hermes.md objective/status/done and added rules 10-11 (keep the brakes fail-safe; wheels
off the ground for first power-up). Next: fill in docs/base-survey.md, starting with the motor wire
count.

2026-11-01 (rev 3) — Mike clarified: he KEPT the whole factory electronics package off the chair
(R-net power module, joystick module, cables — all boxed). He also owns dual brushed motor
controllers rated 35 A and several ESP32s. Rewrote docs/constraints.md rev 3 around TWO paths:
Path A = drive the factory R-net over CAN at 125 kbps (open-source open-rnet/can2RNET exists, with a
published Permobil M3 precedent) — preferred, because the factory power module already implements the
brake sequencing, current limiting and speed limiting we would otherwise write ourselves on a 300 lb
machine; hardware cost ~$5-15 (ESP32 + CAN transceiver) since the ESP32s are owned. Path B = Mike's
35 A brushed controllers + ESP32, only valid if the motors are brushed, and it puts every safety
behaviour on us. Flagged that 35 A is below the motors' ~120 A peak — measured current needed before
committing. Phases re-cut: 1a = bench-test the boxed R-net (an afternoon, ~$15, decides the whole
architecture), 1b = drive safely under our control, 2 = following. Hired the Paperclip worker
'Cart Builder' (id 7ba54b06-496c-44e4-9c95-eb5653fc8998, cwd this project, idle/healthy).
Next: the motor photos, then docs/rnet-bench-test.md.

2026-11-10 (rev 4) — PATH A IS GONE. Mike no longer has the original R-net electronics (power module,
joystick, cables). They were kept and boxed at rev 3; they are gone now. Nothing is left to bench-test,
so Phase 1a and docs/rnet-bench-test.md are cancelled — and with them the option of letting the factory
power module carry the brake sequencing, current limiting and speed limiting. Those behaviours are now
entirely ours to design on a 300 lb machine. That is the real cost of this change, and the reason the
brake write-up is now a gate rather than a nice-to-have. Path B is the only path left, and it is
conditional on the motors being BRUSHED: still unknown, and the one fact everything now waits on
(2 heavy conductors = brushed; 3 plus a ~5-wire bundle = brushless). The rev 3 caveat stands and gets
more pointed without the factory module: Mike's 35 A controllers sit well below the motors' ~120 A
peak, so measured current matters before anything is bought. Updated .hermes.md (Status, next step,
brake consequence) and docs/base-survey.md §4 (R-net rows now read "gone") and §6.
Next: what base hardware survives, and the motor wire count from the survey §2.

2026-11-10 (later) — Nine photos of the base arrived (taken 26–27 Oct 2018, out of OneDrive on the
datapool; copied to output/base-survey/). They show the stripped chassis and give real §1/§3/§4 facts:
KENDA knobby tyres, coil-over suspension on the caster arms, the PUSH/PULL free-wheel lever, the
battery box out of the chair, and a "CE-300 CHANGBAO PVC 105C VW-1" cable marking. They do **not**
answer §2 and cannot: the conductors are inside the sheath, so no count is visible in any shot. An
automated reading offered "typically two heavy conductors" — recorded in base-survey.md §6 explicitly
as an inference and NOT the answer, so nobody later mistakes it for an observation.
Next: the §2 conductor count, which still needs a bare end or a free connector.
