DEV Community

Cover image for Kinewright Wheels: give way, wait, never cut across
Dimon
Dimon

Posted on Originally published at dimonb19a.hashnode.dev

Kinewright Wheels: give way, wait, never cut across

The second Kinewright engine now reads what moves from its own scans, without a camera, and drives near it by declared numbers: it slows, gives way or waits, and never cuts across a moving object's way. When the way stays crossed too long it says so, and this article shows that too.

What this stage adds

  • Tracks from the rover's own scans. Returns the map of what stands still does not explain become clusters, clusters become tracks, and an alpha-beta filter gives each a velocity. No camera, no recognition.
  • A class for each track: moving after three observations at 0.15 m/s or more, static after twelve below it, two seconds. A moving track's returns stay out of the map, so a mover leaves no trail.
  • A mover that stops long enough enters the map, and the route goes round it like any wall.
  • A course for each moving track: constant velocity, two seconds ahead in steps of 0.1 s, with a radius that widens with time.
  • A caution policy with its numbers. The caution field reaches 1.0 m plus two seconds of the track's approach; inside it the shell drives at most 0.25 m/s over the floor.
  • A course check: no step of the shell enters a moving object's predicted way. When every candidate would, the shell gives way at rest, out of the course, and waits.
  • A clear course for five decisions, half a second, ends the wait. After 15 s it stops waiting and says so.
  • A step aside from an oncoming mover: a crab on the four-wheel-steer shell, an S on the front-steer shell.
  • The route planner reads the courses as costs. It replans when a course crosses the route, and again when the mover has passed, so the route comes back straight.
  • Every decision is a recorded input. A drive replays byte for byte; the decisions are integers, identical on two machines of different architectures.

How it decides

  • Ten times a second it reads each moving track's course over the next two seconds against the shell's own candidate paths.
  • A candidate whose outline meets a course in its window is refused; a candidate inside a field is held to the lower speed.
  • A track that falls from moving to static keeps a disc where it rests until the map holds its cells.
  • People are moving objects and nothing else: no field, record or word says a track is a person.

Numbers from this stage

  • On twenty fixture runs, both shells: each ends as declared, with no overlap on any tick, no course crossed and no tick above the lower speed inside a field; 31 yields in all.
  • With the reactive skill alone, a mover crossing ahead: the least distance is 467 and 480 mm, 541 and 512 mm while moving.
  • In the drives shown: on the crossing, the shell gives way at 6.5 s and reaches the goal at 19.4 s; when the mover stops in its way, the route goes round it at 11.1 s and arrives at 29.5 s; under a patrol that keeps crossing a corridor, it waits from 19.2 s and stops waiting at 34.3 s.

What it does not claim, and why

What you see or might expect Why Changes at
It hesitates near a mover; the route switches while it waits The plan is in space only, and a course is a straight line two seconds long The next stage: a plan in space and time
A paused mover is a disc, not a course Constant velocity: a pause has no direction The next stage
A wall seen at a grazing angle can read as moving, and the shell waits on a free road A face inside a map cell, first seen at a grazing angle, passes both tests for an unexplained return; recorded, not repaired The next stage's perception
It knows exactly where it is The rays are placed by the engine's own pose, not an estimate The localisation stage
It recognises people People are moving objects; no person is recognised Never claimed
A yield here is safe A yield is what declared rules chose on tracks from models of declared sensors Never claimed

Two assumptions sit under everything here, as before: the honesty bound of the kinematic model and the tip threshold, both announced when crossed and never simulated. This stage adds no third one; the motion stays kinematic, and nothing is measured.

Top comments (0)