Embedded Files
🎯Objectives
🎯Objectives
Develop a decentralized multi-robot coordination framework using ROS2 Nav2 and DDS networking on TurtleBot4 Lite and Standard platforms.
Enable simultaneous ergodic trajectory tracking across multiple agents with synchronized localization and control.
Achieve high reliability (>95%) and low-latency (<50 ms) communication in distributed navigation without a central controller.
Establish a scalable foundation for heterogeneous robot fleets performing coordinated exploration and coverage tasks.
💡 Motivation
💡 Motivation
Traditional centralized multi-robot systems are limited by single-point failures, network latency, and scalability issues.
With the growing need for collaborative autonomy, robots must coordinate locally and share information without relying on a master node.
ROS2’s DDS communication layer provides a peer-to-peer architecture ideal for decentralized coordination, ensuring fault tolerance and real-time data exchange.
Leveraging this framework on TurtleBot4 Lite and Standard platforms enables scalable testing of ergodic exploration and cooperative navigation in dynamic environments.
⚙️ Approach
⚙️ Approach
1. Communication (DDS Networking & Discovery)
1. Communication (DDS Networking & Discovery)
Initialized the multi-robot network using Fast-DDS Simple Discovery, enabling peer-to-peer topic exchange between TurtleBot4 Lite and Standard robots.
As the system scaled, transitioned to the Fast-DDS Discovery Server configuration (per the TurtleBot4 Networking Guide) to improve scalability, reduce discovery traffic, and stabilize message exchange across multiple namespaces.
Tuned QoS profiles for reliability, history depth, and liveliness to minimize packet loss (<5%) and maintain <50 ms latency between agents.
Decentralized Fast-DDS architecture enabling peer-to-peer communication between TurtleBot4 agents without a central master node.
Decentralized Fast-DDS architecture enabling peer-to-peer communication between TurtleBot4 agents without a central master node.
2. Planner (Trajectory Generation & Coordination)
2. Planner (Trajectory Generation & Coordination)
Began with the NavFn global planner in the ROS2 Nav2 stack to validate baseline waypoint generation and coverage behavior.
Transitioned to the Smac Planner (Hybrid-A*), which provided smoother curvature paths, improved obstacle avoidance, and kinematically feasible trajectories for differential-drive motion.
Integrated an ergodic trajectory generator, partitioning the workspace into subregions assigned dynamically to each TurtleBot for balanced coverage.
Each robot independently executed its assigned path within Nav2, while trajectory state updates were synchronized through DDS to maintain decentralized coordination.
Comparison of path generation across 2D A*, Hybrid-A* (Smac), and State Lattice planners. Hybrid-A* achieves smoother curvature and kinematically feasible trajectories ideal for differential-drive robots.
Comparison of path generation across 2D A*, Hybrid-A* (Smac), and State Lattice planners. Hybrid-A* achieves smoother curvature and kinematically feasible trajectories ideal for differential-drive robots.
3. Controllers (Motion & Coordination Layer)
3. Controllers (Motion & Coordination Layer)
Initially tested the DWB (Dynamic Window Approach) controller, which provided fast local obstacle avoidance but led to oscillatory velocity commands and unstable heading behavior during multi-robot coordination.
Experimented with the Graceful Controller, which improved turning smoothness but produced jerky linear velocity transitions and poor performance under asynchronous updates between agents.
Transitioned to the Regulated Pure Pursuit (RPP) controller, which offered smooth curvature tracking, consistent velocity regulation, and stable convergence on dynamic ergodic paths.
Tuned RPP parameters (lookahead distance, curvature velocity scaling, and rotation threshold) separately for Lite and Standard platforms to ensure uniform trajectory following across the fleet.
TurtleBot4 Control & Coordination Framework
TurtleBot4 Control & Coordination Framework
4. SLAM & Localization
4. SLAM & Localization
Utilized Adaptive Monte Carlo Localization (AMCL) for decentralized pose estimation across each robot namespace.
Shared map frames selectively to maintain global spatial awareness without overloading the DDS network.
Achieved <5 cm localization drift across both Gazebo simulation and real-world TurtleBot4 runs, validating consistency between virtual and physical tests.
AMCL localization and costmap inflation layers visualized in RViz during decentralized navigation.
AMCL localization and costmap inflation layers visualized in RViz during decentralized navigation.
📊 Key Findings
📊 Key Findings
Transitioning from Fast-DDS Simple Discovery to the Discovery Server reduced discovery latency by 40% and eliminated topic dropouts beyond three agents, achieving over 95% message reliability and sub-50 ms latency across four TurtleBots.
Replacing the NavFn planner with Smac Planner (Hybrid-A*) produced smoother curvature paths and reduced computation time by ~35%, enabling simultaneous, collision-free ergodic coverage for 2–4 robots.
The DWB and Graceful Controllers caused oscillatory velocity commands and unstable transitions under distributed updates, while Regulated Pure Pursuit (RPP) maintained smooth curvature tracking and <3 cm inter-robot spacing error after tuning for Lite and Standard platforms.
AMCL-based localization achieved <5 cm positional drift and >90% map consistency, maintaining global spatial alignment across namespaces without overloading the DDS network.
🌍 Impact
🌍 Impact
Demonstrated a scalable, fault-tolerant multi-robot coordination framework capable of reliable decentralized navigation using ROS2 Nav2 and DDS.
Established a stable communication backbone for heterogeneous fleets, enabling consistent coordination across TurtleBot4 Lite and Standard platforms.
Validated the Smac Planner + RPP combination as a robust baseline for distributed trajectory tracking and cooperative exploration tasks.
Provided a transferable framework for multi-agent autonomy, adaptable to aerial, ground, or hybrid robotic systems in research and field deployment.
Laid the groundwork for future integration of ergodic coverage, formation control, and cooperative SLAM under a unified decentralized architecture.
Page updated
Google Sites
Report abuse