RoboRAN: A Unified Robotics Framework for Reinforcement Learning-Based Autonomous Navigation

TL;DR

RoboRAN presents a modular, cross-domain RL navigation framework that decouples robot and task definitions to enable unified training, evaluation, and deployment across terrestrial, aquatic, and microgravity platforms. It introduces four navigation tasks, a standardized reward formulation with domain randomization, and a deployment stack that bridges Isaac Lab training to ROS2-based real-robot control. Through simulation and field experiments on three heterogeneous robots, RoboRAN demonstrates effective sim-to-real transfer, consistent training efficiency, and robust cross-domain performance, while identifying heading alignment and inertia-related gaps as opportunities for improvement. The open-source framework provides a practical foundation for scalable, reproducible RL navigation research and paves the way for multitask learning and transfer-learning approaches across diverse robotic morphologies.

Abstract

Autonomous robots must navigate and operate in diverse environments, from terrestrial and aquatic settings to aerial and space domains. While Reinforcement Learning (RL) has shown promise in training policies for specific autonomous robots, existing frameworks and benchmarks are often constrained to unique platforms, limiting generalization and fair comparisons across different mobility systems. In this paper, we present a multi-domain framework for training, evaluating and deploying RL-based navigation policies across diverse robotic platforms and operational environments. Our work presents four key contributions: (1) a scalable and modular framework, facilitating seamless robot-task interchangeability and reproducible training pipelines; (2) sim-to-real transfer demonstrated through real-world experiments with multiple robots, including a satellite robotic simulator, an unmanned surface vessel, and a wheeled ground vehicle; (3) the release of the first open-source API for deploying Isaac Lab-trained policies to real robots, enabling lightweight inference and rapid field validation; and (4) uniform tasks and metrics for cross-medium evaluation, through a unified evaluation testbed to assess performance of navigation tasks in diverse operational conditions (aquatic, terrestrial and space). By ensuring consistency between simulation and real-world deployment, RoboRAN lowers the barrier to developing adaptable RL-based navigation strategies. Its modular design enables straightforward integration of new robots and tasks through predefined templates, fostering reproducibility and extension to diverse domains. To support the community, we release RoboRAN as open-source.

Figures (10)

• Figure 1: RoboRAN supports easy development of RL-based navigation tasks across a diverse set of robots. The five robots shown are Kingfisher, Floating Platform, Turtlebot2, Leatherback, and Jetbot. All have been implemented in simulation, while the first three have also been evaluated and demonstrated in real-world environments.
• Figure 2: RoboRAN framework: the navigation tasks and Simulation Robots modules, along with a selected RL library of preference, serve as only inputs needed for our Environment Manager to train a policy in simulation, providing a readily available network for deployment on the real analog of the chosen robot.
• Figure 3: Learning curves showing rewards (mean $\pm$ std) over 5 seeds per robot, compared based on task.
• Figure 4: Simulation results across robots and tasks. Performance comparisons for GoToPosition, GoToPose, GoThroughPositions, and TrackVelocities tasks. (a) All robots for GoToPosition. (b, c) FloatingPlatform and Turtlebot2 on GoToPose (distance, heading). (d) Number of goals achieved in GoThroughPositions (all robots). (e) Goals distribution over 4096 parallel evaluation environments (all robots). (f) Linear velocity error in TrackVelocities (all robots).
• Figure 5: Field test results for navigation tasks. Performance evaluation for GoToPose (FloatingPlatform, Turtlebot2), GoToPosition (all robots), and GoThroughPositions (all robots).