Jelly battery: report charging from 27.1 V, not 27.6 V

[shawnpatel]

Problem

The control panel showed Jelly at a resting 100 % while it was on the charger. The lift board read 27.5 V, below the old CHARGING_VOLTS = 27.6, and every voltage above the resting curve's 26.66 V top clamps to 100 %. The charger only reaches its 29.2 V absorption voltage at the very end of a charge, so the old threshold missed almost all of it.

Measurements (Jelly, lift board ADC)

State	Rail
On the charger, nearly full	27.48–27.53 V
Full, unplugged	27.00–27.01 V

ADC noise between samples is ~10 mV.

Change

• CHARGING_VOLTS 27.6 → 27.1 V, with new CHARGING_CLEAR_VOLTS = 27.05 V: BatteryEstimator tracks charging with hysteresis so it doesn't flicker near the threshold, and still treats the charger connecting/leaving as a step instead of averaging across it.
• Docs (robot.battery docstring, docs/api/robot.mdx) updated with the measured values and the limitation.

Known limitation

Early in a charge, a low pack can sit under 27.1 V and show a resting (too high) percentage — voltage alone can't separate that from a full pack at rest. Detecting the step when the charger is plugged/unplugged would close that gap, but needs a recording of that step on the real board first. The full-pack margin (27.0 vs 27.1 V) comes from one reading; if a full pack is seen reading above 27.1 V at rest, raise the threshold.

Testing

• uv run pytest tests/test_battery.py tests/test_jelly_link.py — 62 passed (new: bulk-charge voltage reads charging, measured full-rest 27.0 V does not, hysteresis, reset clears charging).
• Checked estimate_battery against the live readings: 27.01 V → not charging, 27.5 V → charging.
• ruff check / format clean.

🤖 Generated with Claude Code

---

Note

Low Risk
Changes only Jelly battery telemetry thresholds and smoothing; a full pack reading above 27.1 V at rest could still false-trigger charging, which is documented and narrow.

Overview
Fixes the control panel showing 100 % at rest while Jelly was on the charger: bulk charge often sits around 27.5 V, below the old 27.6 V threshold, so charging stayed false even though percent clamped to 100 %.

CHARGING_VOLTS drops 27.6 → 27.1 V (Jelly-measured gap between ~27.0 V full/unplugged and ~27.5 V on the charger). Adds CHARGING_CLEAR_VOLTS = 27.05 V so BatteryEstimator keeps charging latched until the rail falls below the clear level, avoiding flicker near the threshold. Charger plug/unplug is still treated as a voltage step (not smoothed). Smoothed BatteryStatus now uses the estimator’s hysteresis flag instead of re-deriving charging from raw threshold on each sample.

Docs (battery module docstring and docs/api/robot.mdx) note the new thresholds, that percent is meaningless while charging, and that early bulk charge can still read as resting if the rail stays under 27.1 V. Tests cover bulk-charge detection, full resting pack, hysteresis, and reset behavior.

^{Reviewed by Cursor Bugbot for commit 7810101. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
axol	Ready	Preview	Sep 23, 2026 8:51pm UTC