Skip to main content

Nav2: Path planning for bipedal humanoid movement.

Heading Breakdown​

Nav2: Path planning for bipedal humanoid movement adapts the standard navigation stack for walking robots. Nav2 is the second generation of the ROS Navigation stack, built on behavior trees. Path planning is the calculation of a trajectory (list of points) to a goal. Bipedal humanoid movement adds a layer of complexity: unlike a wheeled robot, a humanoid can step over small obstacles but cannot turn in place instantly without stepping. The importance is mobility; a robot that can see but cannot move is a statue. Real usage involves tuning the DWB (Dynamic Window Approach) controller to generate velocity commands that match the Unitree G1's walking gait frequency. An example is setting a "stair cost" in the costmap so the robot prefers the ramp but will take the stairs if necessary. This is key for upgradable systems where we might upgrade the legs to be more agile, requiring a retuning of the planner.

(Note: Sidebar refers to Sim-to-Real, but per mapping, we cover Nav2 here).

Training Focus: Navigation Semantics​

We focus on safety.

  • Costmaps: Layers of danger (inflation layer, obstacle layer).
  • Recovery Behaviors: What to do when stuck (e.g., "backup", "spin").

Detailed Content​

The Nav2 Architecture​

  • Planner Server: Global path (A*, Dijkstra).
  • Controller Server: Local path following (DWB, MPC).
  • Behavior Tree: The logic (Sequence, Fallback).

Bipedal Constraints​

  • Footprint: A humanoid's footprint changes as it walks. We approximate it with a radius.
  • Sway: Humanoids sway side-to-side. The planner must account for this to avoid hitting doorframes.

Industry Vocab​

  • Holonomic: Can move in any direction (humanoids are pseudo-holonomic).
  • Voxel Grid: 3D representation of obstacles.
  • BT (Behavior Tree): XML logic flow.

Code Example: Custom Behavior Tree​

<!-- Defensive Behavior Tree -->
<root main_tree_to_execute="MainTree">
  <BehaviorTree ID="MainTree">
    <RecoveryNode number_of_retries="6" name="NavigateRecovery">
      <PipelineSequence name="NavigateWithReplanning">
        <RateController hz="1.0">
          <ComputePathToPose goal="{goal}" path="{path}" planner_id="GridBased"/>
        </RateController>
        <FollowPath path="{path}" controller_id="FollowPath"/>
      </PipelineSequence>
      <SequenceStar name="RecoveryActions">
        <ClearEntireCostmap name="ClearLocalCostmap-Subtree" service_name="local_costmap/clear_entirely_local_costmap"/>
        <Wait wait_duration="5"/> <!-- Wait for balance to stabilize -->
      </SequenceStar>
    </RecoveryNode>
  </BehaviorTree>
</root>

Real-World Use Case: Crowds​

Navigating a G1 through a crowded hallway. We use the Social Costmap Layer to create "repulsion zones" around people, ensuring the robot doesn't invade personal space while squeezing through the crowd.