← All Projects
Decentralised Mobile Manipulator
GPU-accelerated simulation framework for distributed auction-based task allocation in multi-robot systems
Overview
A comprehensive simulation framework exploring decentralised control architectures for collaborative mobile manipulators. This project investigates distributed auction-based algorithms for coordinating multiple robots in complex task allocation scenarios.
Problem Statement
As multi-robot systems scale, centralised control becomes a bottleneck. This research develops and validates decentralised approaches where robots autonomously negotiate and allocate tasks without global coordination.
Approach
- Implemented distributed auction algorithms for task allocation
- Developed GPU-accelerated simulation for large-scale robot teams
- Created modular architecture for testing various coordination strategies
- Validated performance across different scenario complexities
Key Features
- GPU Acceleration: Leverages CUDA for parallel simulation of hundreds of robots
- Auction-Based Allocation: Robots bid on tasks based on local cost estimation
- Scalable Architecture: Performance tested from 2 to 100+ robot configurations
- Modular Design: Easy to swap coordination algorithms and test scenarios
Technologies
- Python
- CUDA / GPU Computing
- NumPy / SciPy
- Matplotlib for visualisation
Results
Demonstrated significant improvements in task allocation efficiency compared to naive approaches, with near-linear scaling performance as robot count increases.