Gaussian-based Uncertainty-Aware Modelling for Multi-Robot Intelligence

Description

With the concurrent development of computer vision and robotics, Gaussian distribution–based methods for uncertainty-aware spatial and temporal modelling in computer vision (e.g., Gaussian splatting for scene representation, and Gaussian processes for occupancy modelling) have emerged as powerful tools for representing complex models in robotics (e.g., mapping, dynamics, and environmental modeling). A key advantage of Gaussian representations lies in their principled treatment of uncertainty. By jointly encoding state estimates and confidence, Gaussian models provide uncertainty-aware representations that naturally support inference and reasoning. Making them suited for real-world robotic systems operating under dynamic conditions and partial observability. These representations have enabled advances in SLAM, uncertainty-aware learning and decision making, and have been successfully applied in single-robot systems as well as fully connected multi-robot teams operating in real-world missions.

Significant challenges remain, however, when extending these approaches to decentralized or partially connected multi-robot systems (i.e., robot swarms), where the environment and action plan model must be part of a consensus (i.e., a shared and consistent belief across all robots). Achieving consensus in such settings requires accounting for heterogeneous and partial observations, communication constraints, potential faulty behaviours, and pervasive uncertainty in sensing. Gaussian representations inherently support reasoning about uncertainty and confidence, making them a natural foundation for robust consensus mechanisms that enable reliable collective decision making.


This 4-year PhD project aims to design and validate Gaussian models for decentralized multi-robot learning and coordination, with a focus on:

- Developing Gaussian-based representation methods for decentralized multi-robot perception, mapping, and state estimation.
- Designing a consensus framework with a shared Gaussian-based model, potentially leveraging distributed databases with consistancy protocols (e.g., blockchain) and state machines controller deployed on consensus (e.g., smart contracts) to govern consensus updates and decision making.
- Exploiting Gaussian modelling for outlier rejection, reputation management, and reward mechanisms, to guide cooperative learning and robust robot behaviours.
- Evaluating the proposed approaches in simulated and real multi-robot scenarios, with a particular focus on active perception and collaborative SLAM.

What We Offer

- An innovative Ph.D. project at the intersection of robotics, computer vision and machine learning.
- Access to state-of-the-art multi-robot platforms and simulation tools.
- A collaborative research environment with expertise in robotics, computer vision and machine learning.
- Opportunities for international collaboration and publication in leading robotics and AI venues and journals.
- Competitive funding, travel support, and access to modern experimental and computing infrastructure.

Requirements and Conditions

- Master's degree or equivalent in robotics, computer vision, control, computer engineering, or applied mathematics
- Strong background in robotics, probabilistic machine learning, computer vision/perception
- Experience in at least two (2) of:
i) Control of robot swarms,
ii) Gaussian-based uncertainty modelling
iii) Consensus and distributed coordination protocols.
- Programming skills (Python, Solidity, C++ preferred) and familiarity with software tools (ArGoS simulator, ROS, PyTorch, etc.)
- Autonomy, creativity, and motivation to bridge theory and practice in robotics

Please send your application files by email to hanqing.zhao@gel.ulaval.ca
with cc to jflalonde@gel.ulaval.ca

Including the following files:
1. Motivation letter
2. CV
3. BSc and MSc transcripts.