OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Evaluation & Observability

Oceangym

OceanGym is a Python research environment and benchmark for evaluating embodied underwater agents in simulated ocean scenes. It includes perception and navigation tasks, continuous-control actions, and options to modify scenes, objects, environmental parameters, and task settings.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Oceangym — OpenViST project profile in Agent Evaluation & Observability.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use OceanGym when an experiment needs simulated underwater conditions for perception or navigation evaluation, rather than a general-purpose agent framework.

Who it is for

  • Researchers benchmarking autonomous underwater vehicles and other embodied agents.
  • Teams evaluating agent perception with camera or sonar imagery.
  • Developers studying navigation and decision-making in underwater simulation.

Documented uses

  • Compare agent performance across underwater scenes and benchmark tasks.
  • Collect camera-only or camera-and-sonar images for perception experiments.
  • Run navigation decision tasks and inspect evaluation outputs.
  • Customize objects, scenes, and environmental or task parameters.
Before you choose

Requirements & limits

The project repository is publicly accessible. The supplied repository metadata does not identify a license.

  • The documented Unreal Engine decision-task setup is Windows-specific.
  • The repository describes HoloOcean installation and world configuration as part of setup.
  • OceanGym includes both camera and sonar perception workflows.
  • The README contains an IsaacLab conversion path in addition to its Unreal Engine environment.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install OceanGym’s Python package and add a world configuration

Prepare the documented HoloOcean-based OceanGym environment by installing its Python package and world package, then placing a task configuration file in HoloOcean’s worlds directory.

Before you start

  • A cloned OceanGym repository.
  • Python and HoloOcean, as referenced by the installation instructions.
  • A configuration JSON file from asset/decision/map_config or asset/perception/map_config.
  • A Windows environment is shown for the example world-file path.

Steps

  1. From the cloned repository, change to OceanGym_large/client and install the Python package with the command below.

    Documented command
    pip install .

  2. Install the worlds package by running import holoocean followed by holoocean.install("Ocean") in Python.

  3. Copy a JSON configuration from asset/decision/map_config or asset/perception/map_config into HoloOcean’s worlds/Ocean directory; the documented Windows example is C:\Users\Windows\AppData\Local\holoocean\2.0.0\worlds\Ocean.

  4. For first-time use, locate Holodeck.uproject in the engine folder, generate Visual Studio project files, open the project, and follow the README’s Unreal Engine setup guidance before proceeding to its documented world-opening stage.

Expected outcome

The OceanGym Python package and Ocean world package are installed, and a selected task configuration is placed in HoloOcean’s worlds directory.

Guide limitations

  • The README identifies the subsequent Unreal Engine workflow as requiring full UE5 support and states its commands apply to Windows only.
  • The README’s example world path is specific to Windows.
  • This recipe stops at environment preparation; running a task is a separate documented stage.
Documentation 1 ↗
Source-based practical guide

Run OceanGym decision task 4 and evaluate its saved results

Run the documented decision task 4 after preparing the OceanGym environment, then configure and run the evaluation script on its saved memory files.

Before you start

  • The OceanGym environment must first be built using the README’s setup instructions.
  • The experimental code environment is installed with pip install -r requirements.txt.
  • A compatible environment for the documented decision-task workflow; the README says its decision-task commands apply to Windows only because full UE5 support is required.

Steps

  1. Install the experimental code requirements with the command below.

    Documented command
    pip install -r requirements.txt

  2. Start the Unreal Engine project and load the HAIDI map in the demo directory, following the README’s Open the World instructions.

  3. Run task 4 with the command below; use the keyboard instructions for control, or press M to switch to LLM mode. Press Q to exit.

    Documented command
    python decision\tasks\task4.py

  4. After the run, place the generated memory and important_memory files into the corresponding point folders and set the evaluation paths in evaluate.py.

  5. Run the command below to generate evaluation results.

    Documented command
    python decision\utils\evaluate.py

Expected outcome

The task saves logs and memory files in `log/` and `memory/`; the evaluation script saves results under the `\eval\decision` folder.

Guide limitations

  • The README warns that switching to LLM mode may cause temporary lag.
  • The README states that the decision-task commands apply to Windows only and require full support from UE5.
  • The memory and important_memory files must be placed in corresponding point folders and the evaluation paths configured before evaluation.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
OceanGPT/OceanGym · Official homepage title · retained in its source language
How we review sourcesSuggest a correction