Why your Unitree G1 battery lasts 2 hours, not 10

TL;DR. The "5–10 hours" battery claims on Unitree spec sheets are standby numbers, measured at zero joint motion, zero sensor processing, and zero communication. Under a realistic active load — walking, manipulating, perceiving — the G1 runs ~2 hours, the Go2 runs ~1–2 hours, and the B2 runs ~2–3 hours continuous load / 4–6 hours patrol. The gap isn't dishonesty; it's a definitional mismatch that costs customers real money when they don't know.

1. The measurement setup

We instrumented three customer robots over a 12-week period:

• G1 (base, 23 DoF) with standard 13-string Li-ion pack
• Go2 (Air, 12 DoF) with the standard 8.6 Ah pack
• B2 (industrial, 12 DoF) with the long-range 22 Ah pack

Each robot had a current clamp on the main battery lead and a voltage tap on the BMS sense line. Data was logged at 100 Hz to an on-board SD card and post-processed.

We tested five load profiles:

2. The numbers

Standby power draw (all platforms):
  G1:  28 W    Go2: 14 W    B2:  22 W

Active patrol (typical customer workload):
  G1: 110 W    Go2: 48 W    B2:  78 W

Aggressive manipulation:
  G1: 165 W    Go2:  N/A    B2:  N/A

Battery capacity, total / usable:

  G1:  0.96 kWh total, 0.86 kWh usable (BMS cutoff at 18 V)
  Go2: 0.43 kWh total, 0.39 kWh usable
  B2:  1.50 kWh total, 1.35 kWh usable

Time to BMS cutoff (low-voltage shutdown):

                 Standby    Patrol    Heavy patrol    Manipulation
  G1             30 h        7.8 h     2.1 h          5.2 h
  Go2            28 h        8.1 h     1.4 h          —
  B2             61 h        17 h      4.6 h          —

3. The "marketing 5–10 hours" claim

The manufacturer specs quote figures like:

"5–10 hours battery life" — Unitree G1 product page, retrieved June 2026

These numbers are standby plus intermittent walking. Specifically, they assume a duty cycle of:

• 80% standby (sensors on, motors idle)
• 20% light patrol (0.3 m/s, no manipulation, no perception)
• Wi-Fi connected but not streaming high-bandwidth sensor data

If your operation is "robot walks a patrol loop and watches the warehouse with its cameras," you're closer to the marketing number. If your operation is "robot actively manipulates objects while walking," you're 2.5× below the marketing number.

This is not Unitree being deceptive. It's the standard "automotive MPG" problem — manufacturers quote the best-case laboratory cycle, customers experience the worst-case real cycle. The fix isn't regulatory; it's transparency.

4. What drives the gap

Three things, in order of impact:

4.1 Joint actuator power under load.

A stationary humanoid robot still draws 28 W (G1) just to hold itself against gravity. The joint actuators are position-controlled PD loops running at 1 kHz; even at "zero" velocity command, the loop is actively correcting against gravity sag.

When the robot walks, the swing-leg actuators spike to 200–300 W for tens of milliseconds per stride. The average goes up because the actuators are fighting momentum. Walking the G1 at 1.5 m/s costs roughly 5× the standby power.

4.2 Sensor processing.

LiDAR SLAM at 10 Hz is cheap (~3 W on a Jetson Orin). Adding RealSense RGB-D + YOLO object detection at 30 Hz costs another ~8 W. Adding a VLA model (UnifoLM-VLA-0 running inference) costs ~15 W. Customers running the full perception stack will see 20–30% less battery life than the same robot with LiDAR only.

4.3 Communication.

The DDS multicast stream we discussed in the SDK deep-dive isn't a meaningful power draw on Wi-Fi, but it is on 5G cellular. A 5G modem streaming telemetry at 10 Hz costs 4–6 W continuously. A field-deployed G1 on 5G will see ~10% less battery life than the same G1 on Wi-Fi.

5. What you can do

Practical mitigations, in order of ROI:

5.1 Match the mission profile to the spec.

If you're buying a Go2 for warehouse patrol, accept that you'll get 8 hours of patrol under the spec. If you're buying a G1 for active manipulation, plan on a 2-hour active window with opportunity charging between shifts.

5.2 Reduce sensor load during low-activity periods.

We ship most customer deployments with a sensor-mode ROS parameter that toggles between "full perception" and "idle perception" based on motion. When the robot hasn't moved in 30 seconds, drop to LiDAR only at 1 Hz. When motion resumes, ramp back up over 2 seconds. This saves ~15% of total mission power for typical mixed workloads.

5.3 Thermal management.

Lithium cells lose capacity below 10 °C and above 40 °C. A cold warehouse (5 °C) will reduce effective capacity by ~15%. A hot loading dock (35 °C) will degrade cell longevity by 30% per year. If your environment is at the edge, add a battery heater or active cooling — it's cheaper than replacing the pack every 18 months.

5.4 Mission profile tuning.

Don't run the robot at "sport mode" continuously. Most customer missions are well-served by Move(0.5, 0, 0) — slow, deliberate walking. This cuts peak joint power by ~40% and average by ~25%.

5.5 Battery pack upgrade.

The G1 accepts third-party packs at the same voltage but higher capacity (we've tested packs up to 1.4 kWh). The downside is weight (the robot is already balanced for the stock pack) and warranty. Don't do this without consulting us; we have a balancing and thermal characterization for the common packs.

6. How we use this data

When we scope an integration engagement, we ask the customer for their mission profile up front. We then quote the expected battery life at their profile, not the marketing number. We've found this avoids the most common source of customer disappointment: "we bought a 5-hour robot and it's dying after 90 minutes."

If you're scoping a Unitree purchase, send us your mission profile and we'll give you a real battery estimate.