This models the way mobile inspection robots (the kind Boston Dynamics' Spot ships to plants, construction sites and warehouses) actually get used: not roaming freely, but walking a fixed loop of checkpoints, pausing at each one to run a sensor payload (thermal camera, gas sniffer, acoustic mic), then moving on. The four-legged gait is approximated by oscillating each leg's foot in a phase-offset trot cycle so the body bobs and the legs swing believably. Crates scattered on the floor act as obstacles: at low autonomy the robot walks a rigid straight line and can be seen brushing past them; at high autonomy it steers a wider berth around anything in its path, the way an onboard planner replans locally around the fixed route.
P(anomaly at checkpoint) = sensitivity · siteRisk(checkpoint)
battery(t+dt) = battery(t) − drain·dt·(1 + 0.4·speed)
avoidance_radius ∝ autonomy
- Patrol speed — how fast the robot walks the loop; faster patrols cover the site more often but burn battery quicker and shorten dwell time at each checkpoint.
- Autonomy level — how far the planner steers around obstacles versus following the mapped route rigidly; low autonomy is closer to a teleoperated or pre-scripted run, high autonomy approximates onboard SLAM-based replanning.
- Sensor sensitivity — the chance a given checkpoint's scan flags something (a hot bearing, a gas leak, a stray value); real deployments tune this against false-positive cost.
- Battery drain rate — how fast the pack depletes; when it nears empty, send the robot back to its charging dock rather than stranding it mid-route.
Real-world relevance: this is the actual operating pattern behind industrial inspection deployments — a robot doesn't replace a human's judgment, it walks the same repeatable checklist a technician would, at any hour, and only surfaces a checkpoint for a person to review when a reading crosses a threshold.