Skip to content

🤖 Nav2 Baseline

CostNav integrates the ROS2 Navigation Stack 2 as a rule-based baseline for comparison with learning-based approaches. For IL baselines, see IL Baselines.

🚀 Quick Start

make run-nav2
# Defaults: NUM_PEOPLE=20 SIM_ROBOT=segway_e1 FOOD=True TUNED=True AMCL=False

# Then run ONE of:
make start-mission    # single mission
make run-eval-nav2    # batch evaluation

Supported Robots

Robot Command
Nova Carter SIM_ROBOT=nova_carter make run-nav2
Segway E1 SIM_ROBOT=segway_e1 make run-nav2

Architecture

Nav2 Architecture
Nav2 architecture: BT Navigator Server orchestrates Controller, Planner, Behavior, and Smoother servers.

Features

  • Full Nav2 integration with Isaac Sim
  • Mission orchestration with automated start/goal sampling from NavMesh
  • Two localization modes: AMCL (realistic) and GPS-only (ground truth)
  • Tuned and untuned Nav2 parameter sets per robot
  • Cost model integration: energy, distance, time, collision, and food spoilage tracking

Localization Modes

# GPS mode (default — ground truth localization)
make run-nav2

# AMCL mode (realistic — particle filter localization)
make run-nav2 AMCL=True

Each robot has tuned and untuned parameter files:

costnav_isaacsim/nav2_params/
├── segway_e1/
│   ├── navigation_params_tuned_true.yaml   # Tuned for CostNav
│   └── navigation_params_tuned_false.yaml  # Default Nav2 params
├── nova_carter/
│   ├── navigation_params_tuned_true.yaml
│   └── navigation_params_tuned_false.yaml
└── nav2.launch.py

# Use tuned parameters (default)
make run-nav2 TUNED=True

# Use default Nav2 parameters
make run-nav2 TUNED=False

Evaluation

# Run batch evaluation (requires running nav2 instance)
make run-eval-nav2

# Custom timeout and mission count
make run-eval-nav2 TIMEOUT=20 NUM_MISSIONS=100

See Evaluation for detailed metrics and the unified eval script.