CostNav

CostNav :robot:

A cost-driven navigation benchmark for sidewalk robots, built on Isaac Sim.

---

:dart: Overview

CostNav supports a wide range of robot platforms and diverse outdoor environments, and evaluates navigation policies with a unified cost model that captures SLA compliance, operational cost, profitability, and break-even time.

The toolkit enables scalable variation in robots, payloads, maps, and cloud-inference settings, and supports both learning-based and rule-based navigation stacks—making it easy to prototype and compare cost-aware policies without manual tuning for each scenario.

:star2: Highlights

• :moneybag: Business-first benchmark: Policies are evaluated not only on navigation success but also on their operational impact, including robot safety, SLA compliance, profitability, and break-even time—metrics directly tied to real-world deployment.
• :world_map: Diverse environment suite: CostNav provides a set of tasks that span urban, suburban, rural, wild, port, and orchard-style maps, all using the COCO delivery robot with mixed observation (vector + RGB-D) pipelines for consistent evaluation.
• :rocket: Roadmap-ready: Hooks are in place to compare learning vs. rule-based stacks, switch between on-device and cloud inference, and study cost-aware reward shaping.

:movie_camera: Simulation Overview

Environments

Scenario	Description
:cityscape: Sidewalk	City-scale sidewalk map featuring crosswalks, curbs, planters, and other street furniture, delivered via Omniverse USD assets for reproducible layouts.

Agents

Agent	Description
:truck: COCO Robot	Four-wheeled sidewalk courier platform from coco_robot_cfg.py with configurable drive models, cameras, and LiDAR for learning or rule-based controllers.

:chart_with_upwards_trend: Cost Model

graph LR
    A[Navigation Policy] --> B{Simulation}
    B --> C[Task Metrics]
    B --> D[Business Metrics]
    C --> C1[Success Rate]
    C --> C2[Collision Rate]
    D --> D1[SLA Compliance]
    D --> D2[Operating Cost]
    D --> D3[Profitability]

:book: Documentation

:zap: Getting Started

• Quick Reference: Essential commands and configurations
• Architecture Overview: Understand the codebase structure and how components work together
• Environment Versions: Learn about v0, v1, and v2 environments and when to use each

:brain: Core Concepts

• MDP Components: Deep dive into observations, actions, rewards, and terminations
• Robot Configuration: Detailed explanation of the COCO robot’s physical properties, actuators, and sensors
• Cost Model: Understanding business metrics: SLA compliance, operational costs, and profitability

:mortar_board: Guides

• Training Guide: Complete guide to training navigation policies with RL-Games and other frameworks

:books: Reference

• API Reference: Auto-generated API documentation for all modules
• Scripts Reference: Comprehensive reference for all training, evaluation, and utility scripts
• Food Delivery Business: Industry data and economics for food delivery robots