How internrobotics/navdp works
NavDP is a mapless visual-navigation policy and IsaacSim/IsaacLab evaluation platform, with related baseline policies and the X-NavDP reinforcement-learning extension. The main workflow is to run a navigation server, send it task inputs over HTTP from a simulation evaluation, and track the resulting plans with a controller. The repository also contains X-NavDP training environments and policy components. Some server, evaluation, and training files are unsampled, so links to them rely on README documentation or are omitted where wiring is not shown.
Main entry points: eval_nogoal_wheeled.py Wheeled Evaluation · wheeled_tasks.py Navigation Task Configs · worker.py Training Worker · reward_utils.py Task Rewards
Components 24
Simulation Benchmark
- Wheeled Evaluation
eval_nogoal_wheeled.py - Navigation HTTP Client
utils_tasks/client_utils.py - MPC Tracking
utils_tasks/tracking_utils.py - Evaluation Visualization
utils_tasks/visualization_utils.py
NavDP Policy
- NavDP Server
baselines/navdp/navdp_server.py - Navigation Policy
baselines/navdp/policy_network.py - Visual Backbone
baselines/navdp/policy_backbone.py - Depth Estimator
baselines/navdp/depth_anything/depth_anything_v2/dpt.py - DINOv2 Encoder
baselines/navdp/depth_anything/depth_anything_v2/dinov2.py
X-NavDP Training
- RL Environment
baselines/x-navdp/src/environment/env_wrapper.py - Navigation Task Configs
baselines/x-navdp/src/environment/wheeled_tasks.py - Task Observations
baselines/x-navdp/src/environment/tasks/observation_utils.py - Task Rewards
baselines/x-navdp/src/environment/tasks/reward_utils.py - Task Events
baselines/x-navdp/src/environment/tasks/event_utils.py - Curriculum State
baselines/x-navdp/src/environment/tasks/curriculum_utils.py - Scene Assets
baselines/x-navdp/src/training/scene_assets.py - Training Worker
baselines/x-navdp/src/training/worker.py - X-NavDP Policy
baselines/x-navdp/src/x_navdp/models/x_navdp_policy.py
Navigation Baselines
- GNM Model
baselines/gnm/gnm_model.py - ViNT Model
baselines/vint/vint_network.py - NoMaD Model
baselines/nomad/nomad_model.py - ViPlanner
baselines/viplanner/plannernet.py - LoGoPlanner
baselines/logoplanner/policy_network.py
Other components
- Evaluator
Connections 18
- Evaluator → Wheeled Evaluation: runs evaluation
README.md - Wheeled Evaluation → Navigation HTTP Client: requests navigation
eval_nogoal_wheeled.py - Navigation HTTP Client → NavDP Server: sends HTTP requests
README.md - NavDP Server → Navigation Policy: serves policy
- Navigation Policy → Visual Backbone: uses backbone
baselines/navdp/policy_network.py - Visual Backbone → Depth Estimator: uses depth model
baselines/navdp/policy_backbone.py - Depth Estimator → DINOv2 Encoder: uses encoder
baselines/navdp/depth_anything/depth_anything_v2/dpt.py - Wheeled Evaluation → MPC Tracking: tracks trajectories
README.md - Wheeled Evaluation → Evaluation Visualization: visualizes results
eval_nogoal_wheeled.py - Training Worker → Scene Assets: checks scene type
baselines/x-navdp/src/training/worker.py - Training Worker → RL Environment: creates environment
baselines/x-navdp/src/training/worker.py - Training Worker → Task Rewards: selects reward
baselines/x-navdp/src/training/worker.py - Navigation Task Configs → Task Observations: configures observations
baselines/x-navdp/src/environment/wheeled_tasks.py - Navigation Task Configs → Task Rewards: configures rewards
baselines/x-navdp/src/environment/wheeled_tasks.py - Navigation Task Configs → Task Events: configures events
baselines/x-navdp/src/environment/wheeled_tasks.py - Task Events → Curriculum State: uses curriculum state
baselines/x-navdp/src/environment/tasks/event_utils.py - Task Rewards → Curriculum State: uses curriculum state
baselines/x-navdp/src/environment/tasks/reward_utils.py - GNM Model → ViNT Model: shares base model