Distributed Autonomous Sensing Systems

The Distributed Autonomous Systems theme centers on the fundamental signal processing challenges of multi-agent sensing systems. The current research focus is on distributed/decentralized processing, probabilistic machine learning, federated learning, resource allocation, autonomous inference through sensor fusion, and Navigation (including localization and synchronization). One of the underlying challenges is to develop algorithms for resource constrained sensor systems, and which are particularly deployed in inaccessible locations. We work closely with various universities (e.g, TU/e, Radboud), research institutes (e.g., TNO, SRON, NLR) and industries (e.g., VSL, KPN, NXP), in the Netherlands and internationally.

 

Techniques

  • Estimation and Detection theory
  • Convex and Non-convex optimization, Mixed integer optimization
  • Statistical signal processing, performance analysis and bounds
  • Distributed signal processing and Distributed systems
  • Parametric/Non-parametric models for complex dynamical systems
  • Dynamics and Kinematics of multi-agent systems
  • Probabilistic machine learning, Sensor fusion

Applications

  • Wireless sensor networks
  • Cooperative and non-cooperative localization and tracking of mobile agents
  • Data and clock synchronization, multi-sensor calibration
  • Autonomous navigation, Environment perception, Path planning and Detect and Avoid
  • Multi-agent systems e.g., aircrafts, rovers, satellite arrays and drone swarms

Research Themes

  • Signal processing for Distributed inference: Sensor systems consist of numerous nodes with only local communication capabilities. Challenges include localization of the nodes, clock synchronization, low-power communication protocols, and distributed estimation/detection algorithms (e.g., consensus, ADMM, PDMM), where local estimates are combined to form global parameter estimates.
  • Signal processing for Autonomous inference: A single system consisting of multi-modal sensors (e.g., IMU, Camera, LIDAR, Radar) aims to autonomous perform desired tasks with minimal control from other actors/agents/humans e.g., environmental perception, path planning, detect and avoid, and self-calibration. How can the diverse streams of data from various sensors (e.g., continuous, discrete, intermittently available, block datasets) be combined in a prudent way for optimal statistical inference?
  • Signal processing for multi-agent mobile systems:  Consider a network of mobile agents (e.g., automotive, satellites, rovers, drones), which cooperatively perform a given task e.g., platooning, collaborative path planning and formation flying. How can we exploit the underlying physics, and optimally combine information of kinematics and dynamics of these multi-agent systems to perform these tasks?

Projects under this theme

Delft Sensor AI Lab

Distributed AI for sensor networks

Cooperative Relative Navigation of Multi-agent Systems

Develop algorithms for multi-target position, time and orientation tracking in a mobile network of multi-agent systems

Automotive Intelligence for Connected Shared Mobility

Architectures for embedded intelligence and functional virtualization for connected and shared mobility using trustworthy AI

Distributed Artificial Intelligent Systems

Running existing algorithms on vastly distributed edge devices

Airborne data collection on resilient system architectures

Develop algorithms to realize efficient, robust, cost-effective perception and control for autonomous navigation of drones

PIPP OLFAR: Breakthrough technologies for Interferometry in Space

Combine multiple satellites into one single scientific instrument: a radio telescope in space

History

Low-frequency distributed radio telescope in space

Below 15 MHz, the ionosphere blocks EM signals from the sky. Therefore, can we design a radio telescope in space, using a swarm of inexpensive nano-satellites? Accurate localization and clock recovery is important.