Multi-Agent Systems and Autonomous Vehicles

Current Developments in Multi-Agent Systems and Autonomous Vehicles Research

The recent advancements in the field of multi-agent systems and autonomous vehicles have shown a significant shift towards more data-driven, adaptive, and robust control strategies. This report highlights the general direction that the field is moving in, focusing on innovative approaches that advance the state-of-the-art.

Data-Driven and Adaptive Control

One of the prominent trends is the integration of data-driven methodologies into control systems. Researchers are increasingly leveraging machine learning and reinforcement learning techniques to create controllers that can adapt to unknown environments and varying system dynamics. This approach is particularly useful in scenarios where traditional model-based control methods fall short due to the complexity and unpredictability of real-world conditions.

For instance, learning-based controllers are being developed to handle extreme variations in quadcopter dynamics, eliminating the need for precise model estimation. These controllers use a combination of imitation and reinforcement learning to adapt swiftly to diverse conditions, as seen in the "A Learning-based Quadcopter Controller with Extreme Adaptation."

Infrastructure-Less and Distributed Systems

There is a growing interest in infrastructure-less solutions for localization and control in multi-agent systems. Ultra-wideband (UWB) technology is emerging as a viable alternative to traditional methods that require fixed external structures. Innovations in UWB-based relative localization allow for more flexible setups, enabling robots to adapt their positions to minimize localization errors without relying on a static anchor.

Distributed control algorithms are also gaining traction, particularly in scenarios involving large-scale interconnected networks. These algorithms leverage the interconnection topology to break down the network analysis into manageable subsystems, enhancing scalability and robustness. The "From Data to Control: A Formal Compositional Framework for Large-Scale Interconnected Networks" exemplifies this approach by introducing a compositional data-driven methodology for decentralized safety controller design.

Robust and Probabilistically Robust Planning

Robust trajectory planning remains a critical area of research, especially in the context of autonomous agents operating in uncertain environments. Recent advancements focus on developing probabilistically robust controllers that can handle non-Gaussian uncertainties and nonlinear systems. These controllers transform non-Gaussian chance constraints into deterministic ones, enabling safer and more accurate path planning.

The "Probabilistically Robust Trajectory Planning of Multiple Aerial Agents" introduces a novel method that leverages mixed-trigonometric-polynomial moment propagation to achieve this transformation, ensuring robust and accurate path planning in complex scenarios.

Noteworthy Innovations

Several papers stand out for their significant contributions to the field:

  1. "Data-Driven Cooperative Output Regulation of Continuous-Time Multi-Agent Systems with Unknown Network Topology": This paper introduces a novel approach to compute controller parameters without prior knowledge of the network topology, significantly advancing the field of cooperative control.

  2. "Accurately Tracking Relative Positions of Moving Trackers based on UWB Ranging and Inertial Sensing without Anchors": The integration of a custom Extended Kalman Filter with multidimensional scaling for relative positioning without anchors is a notable innovation in infrastructure-less localization.

  3. "Infrastructure-less UWB-based Active Relative Localization": The active method for infrastructure-less relative localization, which allows robots to adapt their positions to minimize localization errors, represents a significant step forward in multi-robot systems.

  4. "A Learning-based Quadcopter Controller with Extreme Adaptation": The learning-based controller's ability to generalize to unseen quadcopter parameters and adapt to extreme conditions highlights the potential of machine learning in control systems.

  5. "Probabilistically Robust Trajectory Planning of Multiple Aerial Agents": The novel method for handling non-Gaussian uncertainties and nonlinear systems in trajectory planning is a major advancement in robust control.

These innovations collectively underscore the field's progress towards more adaptive, robust, and scalable solutions for multi-agent systems and autonomous vehicles.

Sources

Data-Driven Cooperative Output Regulation of Continuous-Time Multi-Agent Systems with Unknown Network Topology

Accurately Tracking Relative Positions of Moving Trackers based on UWB Ranging and Inertial Sensing without Anchors

Infrastructure-less UWB-based Active Relative Localization

A Learning-based Quadcopter Controller with Extreme Adaptation

Linear Model Predictive Control for Quadrotors with An Analytically Derived Koopman Model

From Data to Control: A Formal Compositional Framework for Large-Scale Interconnected Networks

Probabilistically Robust Trajectory Planning of Multiple Aerial Agents

Aircraft Conflict Detection and Avoidance through Interacting Multiple Model (IMM) Estimation

Scalable Multi-Objective Optimization for Robust Traffic Signal Control in Uncertain Environments

Obstacle-Free Path Planning for Autonomous Drones Using Floyd Algorithm

Distributed Control for 3D Inspection using Multi-UAV Systems

Model Predictive Control For Multiple Castaway Tracking with an Autonomous Aerial Agent

Cooperative distributed model predictive control for embedded systems: Experiments with hovercraft formations

ProxFly: Robust Control for Close Proximity Quadcopter Flight via Residual Reinforcement Learning

Concurrent and Scalable Trajectory Optimization for Manufacturing with Redundant Robots

LiDAR-based Quadrotor for Slope Inspection in Dense Vegetation

Kinodynamic Motion Planning for Collaborative Object Transportation by Multiple Mobile Manipulators

Enabling On-Chip High-Frequency Adaptive Linear Optimal Control via Linearized Gaussian Process

A Contract Theory for Layered Control Architectures

Intent Prediction-Driven Model Predictive Control for UAV Planning and Navigation in Dynamic Environments

NavRL: Learning Safe Flight in Dynamic Environments

Distance-based Multiple Non-cooperative Ground Target Encirclement for Complex Environments

Real-time Planning of Minimum-time Trajectories for Agile UAV Flight

Multi-UAV Pursuit-Evasion with Online Planning in Unknown Environments by Deep Reinforcement Learning

Multirotor Nonlinear Model Predictive Control based on Visual Servoing of Evolving Features

Collision-free time-optimal path parameterization for multi-robot teams

Dashing for the Golden Snitch: Multi-Drone Time-Optimal Motion Planning with Multi-Agent Reinforcement Learning

Built with on top of