zju3dv

zju3dv / habitat-gs

Public

Habitat-GS: A High-Fidelity Navigation Simulator with Dynamic Gaussian Splatting

zju3dv.github.iohabitat-gs computer-vision embodied-ai embodied-navigation gaussian-avatar gaussian-splatting
77
2
100% credibility
Found Apr 22, 2026 at 77 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
C++
AI Summary

Habitat-GS is a high-fidelity simulator extending Habitat-Sim to support photo-realistic Gaussian Splatting scenes and dynamic avatars for embodied AI navigation tasks.

How It Works

1
🔍 Discover Habitat-GS

You find this simulator on GitHub, excited by its promise of lifelike 3D worlds for training navigation agents.

2
⚙️ Set up your simulator

Follow simple steps to create a ready-to-use environment on your computer.

3
📥 Download scenes and avatars

Grab high-quality 3D scenes and animated characters to populate your worlds.

4
🖥️ Explore interactively

Launch the viewer and walk around stunning photo-realistic environments with moving avatars.

5
🚀 Train navigation agents

Use one-click scripts to teach AI agents to find goals, follow instructions, or navigate objects.

6
📊 Test and evaluate

Run evaluations to see your agents succeed in realistic simulations.

🎉 Agents navigate real worlds

Your trained agents now handle complex navigation in high-fidelity scenes, ready for embodied AI research.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 77 to 77 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is habitat-gs?

Habitat-GS is a C++ simulator that extends Habitat with high-fidelity Gaussian splatting rendering for dynamic navigation tasks. It renders photo-realistic scenes—like gsm habitat or gss herbal habitat photos—and drives humanoid Gaussian avatars, while preserving Habitat's navmesh pathfinding and sensor suite. Users get Habitat-Lab integration for training agents on PointNav, ImageNav, ObjectNav, and VLN in realistic environments.

Why is it gaining traction?

Gaussian splatting delivers faster, sharper renders than mesh-based sims, enabling dynamic scenes without sacrificing fidelity. Ready-to-run scripts generate episodes, train DDPPO baselines, and eval VLMs like StreamVLN or Uni-NaVid, plus an interactive viewer and HF dataset of 65 GS scenes. It's a quick upgrade for Habitat users chasing habitat-gsf or habitat gsl realism.

Who should use this?

Embodied AI researchers benchmarking navigation agents on photo-realistic outdoor scenes, like habitat beside gsl medical college layouts. RL teams training PointGoal or ObjectGoal policies, or VLM devs fine-tuning VLN on Gaussian splatting data.

Verdict

Worth trying for high-fidelity navigation sims—docs and pipelines are polished despite 77 stars and 1.0% credibility score. Maturity shows in C++ core and one-click workflows, but verify navmeshes on your scenes first.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.