RT-GuIDE: Real-Time Gaussian splatting for Information-Driven Exploration

Yuezhan Tao,Dexter Ong,Varun Murali,Igor Spasojevic,Pratik Chaudhari,Vijay Kumar

2024-09-27

Abstract:We propose a framework for active mapping and exploration that leverages Gaussian splatting for constructing information-rich maps. Further, we develop a parallelized motion planning algorithm that can exploit the Gaussian map for real-time navigation. The Gaussian map constructed onboard the robot is optimized for both photometric and geometric quality while enabling real-time situational awareness for autonomy. We show through simulation experiments that our method is competitive with approaches that use alternate information gain metrics, while being orders of magnitude faster to compute. In real-world experiments, our algorithm achieves better map quality (10% higher Peak Signal-to-Noise Ratio (PSNR) and 30% higher geometric reconstruction accuracy) than Gaussian maps constructed by traditional exploration baselines. Experiment videos and more details can be found on our project page: <a class="link-external link-https" href="https://tyuezhan.github.io/RT_GuIDE/" rel="external noopener nofollow">this https URL</a>

Robotics

What problem does this paper attempt to address?

### Problems Addressed by the Paper

This paper aims to address the problem of real-time information-driven exploration and mapping by autonomous robots in unknown environments. Specifically, the authors propose a method based on Gaussian splatting to construct information-rich maps and develop a parallelized motion planning algorithm to achieve real-time navigation using Gaussian maps. This approach not only optimizes the photometric and geometric quality of the maps but also enhances the robot's real-time situational awareness.

### Main Contributions

1. **Unified Framework**: A unified framework for online mapping and planning is proposed, using Gaussian distributions to represent the environment.
2. **Approximate Information Gain Metric**: An easy-to-compute information gain metric that can run in real-time on unmanned ground vehicles is proposed.
3. **Gaussian Frontier Computation Method**: A method for computing frontiers in Gaussian splatting-based maps is proposed.
4. **Experimental Validation**: The effectiveness of the proposed framework is validated through extensive experiments in both indoor and outdoor environments.

### Method Overview

1. **Mapping Module**:
- The environment is represented using the 3D Gaussian Splatting (3DGS) method, achieving real-time radiance field rendering through the projection and blending of 3D Gaussian distributions.
- The 3D Gaussian parameters are optimized to improve the quality of scene representation.
- The Gaussian uncertainty in the map is estimated to select the best viewpoints.

2. **Planning Module**:
- High-level Planning: A graph structure is constructed, adding observed and predicted nodes, and the Dijkstra algorithm is used to find the shortest path from the current robot position to the target node.
- Trajectory Generation: An optimized information-gathering trajectory is found while ensuring dynamic feasibility and collision avoidance.

### Experimental Results

1. **Simulation Experiments**:
- Experiments were conducted on the iTHOR dataset in the AI2-THOR simulator, evaluating metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), Learned Perceptual Image Patch Similarity (LPIPS), and Root Mean Square Error (RMSE).
- The results show that the proposed method outperforms or is comparable to baseline methods in all metrics, with significantly faster computation speed.

2. **Real-world Experiments**:
- The algorithm was deployed in two different real environments, tested using the Clearpath Jackal robot.
- Compared to voxel map-based baseline methods, the proposed method excels in map quality and coverage, especially when a larger exploration budget is allowed, significantly improving map reconstruction quality.

### Conclusion

The method proposed in this paper makes significant progress in real-time information-driven exploration and mapping, particularly in the construction and utilization of Gaussian splatting maps. Experimental results validate the effectiveness and superiority of the method, providing new possibilities for further robotic applications.