Embedded Files
🎯 Objectives
🎯 Objectives
Develop a distributed optimal control framework for multi-robot convoys, validated in both MATLAB simulation and ROS2 + Gazebo environments.
Implement iLQR-based controller capable of real-time trajectory tracking and convoy formation under nonlinear vehicle dynamics.
Maintain tight, stable inter-robot spacing while minimizing a quadratic cost over tracking deviation and formation control terms.
Integrate adaptive spacing penalty and coordination terms into the control law to ensure collision avoidance and cooperative maneuvering.
Enable each agent to compute locally optimal actions over a time horizon using neighbor state feedback, achieving decentralized formation control without a central planner.
Demonstrate globally coherent and robust multi-agent coordination, scalable across heterogeneous robot fleets through distributed optimization.
Schematic for the convoy control problem. Lead and follow vehicles are defined with respect to vehicle i. Error e is the Euclidean distance between a vehicle’s desired position (d_ref behind the lead vehicle) and its current location.
Schematic for the convoy control problem. Lead and follow vehicles are defined with respect to vehicle i. Error e is the Euclidean distance between a vehicle’s desired position (d_ref behind the lead vehicle) and its current location.
Fleet response illustrating reduced “accordion effect” when follower feedback is incorporated in the distributed iLQR framework.
Fleet response illustrating reduced “accordion effect” when follower feedback is incorporated in the distributed iLQR framework.
đź’ˇ Motivation
đź’ˇ Motivation
Conventional convoy control strategies—such as leader–follower and rule-based frameworks—often fail to balance key objectives like stability, responsiveness, and energy efficiency, especially when operating in dynamic or unstructured environments.
An optimal control formulation overcomes these limitations by minimizing trajectory tracking error, control effort, and inter-robot deviation within a unified cost-based optimization framework.
Building on this foundation, a distributed nonlinear predictive control architecture allows each robot to compute locally optimal yet globally consistent actions using only neighbor state feedback. This enables robust, adaptive, and scalable convoy coordination, validated in both MATLAB simulations and ROS2 + Gazebo experiments.
⚙️ Approach
⚙️ Approach
Convoy Controller Design:
Convoy Controller Design:
The weighting matrices Q, Q_f​, and R are symmetric and positive definite, tuned to balance tracking accuracy, smoothness, and terminal precision.Â
The weighting matrices Q, Q_f​, and R are symmetric and positive definite, tuned to balance tracking accuracy, smoothness, and terminal precision.Â
The formation cost couples each agent with its immediate leader and follower, enforcing bidirectional spacing constraints that maintain a stable convoy formation without oscillations or collisions.
The formation cost couples each agent with its immediate leader and follower, enforcing bidirectional spacing constraints that maintain a stable convoy formation without oscillations or collisions.
Controller Implementaton:
Controller Implementaton:
đź“ŠSimulation Results:
đź“ŠSimulation Results:
The distributed convoy controller was validated in both MATLAB simulation and ROS2 + Gazebo environments using 2–7 TurtleBot4 robots under varying trajectories and spacing conditions.
Simulation Setup:
Simulation Setup:
No. of robots: 2-7
Sim Environment: Gazebo Harmonic and Matlab
Controller frequency: 20 Hz
Control Source: Reference path or teleoperation (only robot 1)
Communication: DDS topics – /odom, /cmd_vel, /tf.
Control loop timing statistics for the distributed iLQR controller. The controller maintains real-time operation (≤ 0.05 s) up to three robots. Beyond this, loop times begin exceeding the 20 Hz control threshold, indicating increased computational load due to inter-agent coupling and solver complexity growth with fleet size. The observed delay is partly attributed to all robot control nodes being executed on the same machine; in real deployments with distributed computation, this latency would be significantly reduced.
Control loop timing statistics for the distributed iLQR controller. The controller maintains real-time operation (≤ 0.05 s) up to three robots. Beyond this, loop times begin exceeding the 20 Hz control threshold, indicating increased computational load due to inter-agent coupling and solver complexity growth with fleet size. The observed delay is partly attributed to all robot control nodes being executed on the same machine; in real deployments with distributed computation, this latency would be significantly reduced.
🌍 Impact
🌍 Impact
The distributed iLQR framework enables real-time, collision-free convoy coordination with adaptive spacing under dynamic conditions. It demonstrates scalable, computation-efficient multi-robot control applicable to cooperative navigation and autonomous fleet operations.
Page updated
Google Sites
Report abuse