The bug is on the power rail: why robots pass on the bench and fail on the floor

A post published on dev.to lays out the symptom list that most mobile robot teams will recognise: the machine runs correctly on the desk, then on the ground it randomly restarts, motors stop during acceleration, sensors return incorrect data and the wireless link disconnects ^[1]. The author's argument is that the first instinct, to assume a bug in the code, is often wrong, and that the real fault sits in the power system ^[2].

The numbers in the post explain the mechanism. A robot is described as going from low draw to high current within milliseconds ^[3], with illustrative figures of 500mA idle, 3A moving and 8A or more under acceleration ^[4]. That is a 16-fold swing ^[5], against a comparison device that wanders from 100mA to 120mA and back ^[6]. Motors are named as the dominant consumer, with demand spiking on startup, direction change, climbing an obstacle or carrying a heavier load ^[7].

The worked voltage example is the useful part. A pack sitting at 3.8V drops to 3.5V when current steps from 1A to 8A ^[8]. Work backwards from those figures and you get roughly 43 milliohms of source impedance between the cell chemistry and the load ^[9]. That number is not a battery datasheet line. The post attributes sag to cell internal resistance, and separately to small wires, unsuitable connectors and long power paths, noting that even small resistance values matter at high current ^[10][11]. At 8A, 43 milliohms turns about 2.7W into heat inside the pack, the wire and the contacts ^[12].

Every item on the symptom list is downstream of that one event. The post lists microcontroller reset, communication failure, sensor malfunction and unexpected shutdown as consequences of a drop the regulator or MCU cannot ride through ^[13], and says the pattern is common in robot dogs, autonomous vehicles, robotic arms and mobile robots ^[14]. Four defects that look independent in a log file can share a single cause on the rail.

This is also why the usual battery upgrade backfires. Choosing on capacity alone is called out as one of the biggest mistakes in robot design ^[15]; the post's comparison puts a 5000mAh pack with lower discharge capability against a 3000mAh pack with higher discharge capability and concludes the smaller one may perform better ^[16], despite storing 40 per cent less energy ^[17]. Runtime is a capacity question; surviving acceleration is a current question.

Temperature moves the same variable. In the cold, internal resistance rises, available capacity falls and voltage sags faster under load ^[18], which the post flags for outdoor robots, drones and industrial devices ^[19]. A warm bench test says nothing about a cold morning.

What to watch is the rail, not the repository. Measure voltage at the MCU supply pin during the worst acceleration the robot can produce, at a sample rate fast enough to catch an event the source describes as millisecond-scale ^[3]. Divide the voltage drop by the current step; if the answer is tens of milliohms, the next suspects are connector contacts, wire gauge and path length rather than the control loop ^[10][11]. Then repeat the test cold ^[18].

One caveat on provenance: this is a single blog post using illustrative figures, not instrumented data from a named platform ^[20]. The arithmetic is consistent, and the failure mode is checkable on your own hardware in an afternoon, which is the only verification that counts.