Native Navigation

OhhO OS ships a built-in, no-ROS navigation stack (ohho.nav) — the lightweight counterpart to Nav2's core loop, pure standard library, running on any robot with base.drive:

  • Occupancy-grid mapping from the robot's planar scan (perception.scan)
  • Costmap obstacle inflation by the robot's radius
  • A* path planning (8-connected, replans while driving)
  • Frontier exploration — drive into the unknown until the map is complete
  • A closed-loop Navigator with collision-aware stuck detection and goal diversion ("go to the table" stops beside the table)

For full Nav2 / SLAM-toolbox navigation, use the ROS 2 runtime instead — this stack is for the no-ROS path and for robots without a ROS stack.

CLI

# Build and print the map (spin-scan in place)
ohho nav map sim --transport sim:// --world demo

# Navigate to a world-frame point (A* over the live map, replanning)
ohho nav goto sim --transport sim:// --world demo --x 2.0 --y 1.0
# reached at (1.78, 0.84), 0.14 m from goal

# Frontier exploration
ohho nav explore sim --transport sim:// --world demo --frontiers 4

--world demo furnishes the simulator with labeled obstacles (chair, table, plant, box, cup). On a real robot, the map builds from the robot's scan.

Python

from ohho import Robot
from ohho.nav import Navigator

bot = Robot.connect("omnibot")
nav = Navigator(bot)

result = nav.navigate_to(2.0, 1.0)      # NavResult(reached, reason, x, y, …)
print(result)                            # reached at (1.98, 1.02), 0.05 m from goal

for line in nav.explore(max_frontiers=4):
    print(line)

print(nav.map_ascii())                   # '#' occupied · '·' free · ' ' unknown

Deterministic testing: every loop accepts step_fn(dt) — pass the sim's step to advance physics synchronously (no threads, no sleeps).

Behavior details

  • Goal inside an obstacle → the navigator diverts to the nearest reachable free cell and reports goal (adjusted).
  • Stuck against something unmapped → after ~1.2 s without progress it either declares arrival (if already beside the target) or bump-marks the blocking cell and replans around it.
  • No scan capability → the planner is optimistic over unknown space (straight-line A*); mapping simply stays empty.
  • Holonomic bases strafe toward waypoints; differential bases rotate-then-drive.

Agent tools

When the agent runs (ohho agent …), navigation is exposed as tools the LLM can call: navigate_to(x, y) and explore() — see Memory for the perception/recall tools that pair with them.

All docs