Expand description
Long Range Navigator-lite: affordance-scored frontier navigation.
This reproduces the decision core of long-range navigation past the local map: a robot with a limited, occlusion-aware sensor cannot see a distant goal, so it repeatedly picks the frontier (a known-free cell bordering unknown space) that best advances toward the goal, then hands off to a local planner that drives there over the known-free map. As the robot moves it reveals more of the world, re-evaluates frontiers, and eventually sees a path to the goal.
- Sensing is occlusion-aware: cells are revealed only along a clear line of sight within the sensor range, so obstacles cast unknown shadows whose edges become frontiers.
- Each frontier is scored by an affordance combining goal progress, the known-free travel cost to reach it, whether it is in direct line of sight, and how much unknown space it borders (information gain).
- The local-planner handoff is a Dijkstra distance field over the known-free map: the robot follows its gradient toward the chosen frontier for a bounded step budget before re-sensing.
Everything is deterministic for a fixed world and configuration.
Structs§
- Frontier
Choice - One frontier selection, retained for inspection and rendering.
- Frontier
NavConfig - Configuration for the frontier navigator.
- Frontier
NavReport - Result of a frontier-navigation rollout.
- Frontier
NavWorld - A static occupancy world the robot explores.
Enums§
- Knowledge
- Known-map cell state.
Functions§
- simulate_
frontier_ navigation - Drive the affordance-scored frontier navigator across the world.