Gaussian Representations for Robot Autonomy: Challenges and Opportunities

 Description

This workshop explores recent advancements in Gaussian representations for robot perception, presents technological and scientific challenges, and highlights opportunities for research and development in this rapidly evolving area. Perception is a fundamental robotics technology upon which many downstream subsystems rely (e.g., motion planning). However, complex autonomous systems frequently generate distinct representations of sensor data in concurrent processes to meet the demands of real-time operation. This architecture strikes a balance between real-time operation and high-fidelity modeling for size, weight, and power (SWaP) constrained robots, but comes at the cost of redundancy. A further challenge is these intermediate representations may not be directly interpretable by humans, which limits human-in-the-loop operations.

This workshop proposes a paradigm shift in robot autonomy to overcome these challenges. Motivated by recent advancements in Gaussian representations, we investigate how Gaussians may be leveraged as a common processing element for higher-level autonomy tasks, thereby creating a framework that unifies disparate pipelines under a common representation. To remain responsive to the demands of field deployable systems, the workshop will emphasize discussions on the impact on compute, opportunities for parallelization, and development of specialized hardware.

The intended audience for this workshop are researchers working on safe navigation for robot autonomy, developing 3D scene representations, and RGB-D perception. Confirmed speakers for this workshop specialize in leveraging Gaussian representations for these applications.

Important Dates

• Submission Start: Apr 19 2025 11:59PM UTC-0

• Submission Deadline: May 01 2025 11:59PM UTC-0

• Acceptance Notification: June 05 2025 11:59PM UTC-0

• Camera-Ready Deadline: June 10 2025 11:59PM UTC-0

• Workshop Date: June 25th 2025

Submission Instructions

1. Submissions must adhere to the RSS 2025 paper format available here under the section "Paper and Demo format".

2. Reviews will be double-blind, so ensure your paper is appropriately anonymized.

3. Although there is no strict page limit, a length of 4–8 pages (excluding references and supplementary material) is recommended.

4. All papers must be submitted through OpenReview.

5. Accepted papers will be presented as posters, with a select few receiving invitations for spotlight presentations.

Topics Covered

• Continuous Scene Representations 

• Applications of Gaussian Processes in Robotics

• Applications of Gaussian Mixture Models in Robotics

• Applications of Gaussian Splatting in Robotics

• RGB-D Perception

• Collision Avoidance

• Continuous-space Motion Planning

• Robotic Exploration

• Active Sensing

• Active Learning

• Multi-Modal Reasoning