Research‎ > ‎

Cooperative Autonomous Vehicles

Autonomous vehicles are expected to play a key role in future transportation systems, as they can lead to augmented safety, higher productivity, and better mobility efficiency, as well as a positive impact on the environment.

In the last couple of year, we have been seeing several research efforts to develop autonomic vehicle technology, driven by advances in sensing, computing and networking technologies.  On the one hand, autonomous driving on urban roads has seen significant progress in recent years. On the other hand, autonomous Unmanned Aerial Vehicles (UAVs) have received increasing interest to tackle several use-cases from environment monitoring and emergency situations, to relaying communications to isolated areas.

Most of the research effort has been looking at several issues related to the control of a single vehicle, such as the capability to avoid obstacles and perform route planning with minimal user intervention. However, there are more demanding scenarios that require the cooperation between  autonomous vehicles, such as: i) in the case of UAVs, scenarios that require fast reaction (e.g. search and rescue) or formation along environmentally friendly routes; ii) in the case of self-driving cars, scenarios with convoys in highways to reduce fuel consumption, or cooperation to avoid pedestrians.

  • Dynamic missions with changing objectives such as search and rescue for UAVs, of autonomous intersection management for self-driving cars, require fast reaction to new environmental and mission settings, which has an impact on sense-and-avoid as well as route planning strategies.
  • Formation along highways (on the sky or on the ground) require vehicles to “self-organize” and select the most efficient and environmentally friendly routes, making the optimum use of prevailing weather and traffic conditions, bringing efficiency improvements due to lower energy use.

Current Research Issues

A) Cooperative Perception
Cooperative perception allows the exchange of perceived data locally acquired by each autonomous vehicles. Perception is a fundamental function to enable autonomous vehicles, providing the vehicle with crucial information on the moving environment, including location of surrounding obstacles, velocities, and even predictions of their future states. The perception data can either be raw sensor data from radar (e.g. LIDAR), cameras or other sensors, or fused data that contains a list of detected objects and their relative positions and predicted trajectories.

In a real scene, the points that a vehicle collects from the LIDAR are not perfect, due to missing points or inconsistent relative location. The surrounding environment also adds more challenges to the perception, as surfaces may be arbitrary and erratic. Hence, simply adding together the location information from different vehicles is not enough: inconsistent perception (due to the vehicle movement) results can lead to dangerous motion behaviours. Therefore, we need a cooperative perception system, allowing vehicles to have a larger area of understanding, and thereby significantly improve environmental awareness. Fusing perception information can potentially reduce uncertainty, allowing the early detection of possible hazards and thereby allow autonomous vehicles to have a faster response to avoid dangerous accidents.

Key points of this work item are:
  • Usage of Light Detection and Ranging (LIDAR) for perception of obstacles and surround mapping.
  • Exploiting vehicle cooperation as fog nodes to improve perception of position in surrounding environments.
B) Cooperative Planning
Cooperative planning allows autonomous vehicles to coordinate their trajectories to achieve specific .goals. Cooperative planning allows autonomous vehicles to rely on local decisions by properly interacting with other agents and following rules restrictions, generating local objectives, such as change trajectory or change position in relation to other vehicle. For instance, with cooperative planning it will be possible to have multiple autonomous vehicles spreading over multiple trajectories maintaining a pre-designed formation. Cooperative planning is usually built on the top of cooperative perception as vehicles usually exchanges their intended trajectories.

In robotics, formation control for multiple robots have been an active research area for decades. However, such research findings are not suitable for the control of formation of autonomous vehicles. Firstly, vehicles are constrained to move in a structured environment (a multi-lane road or a defined aerial route). Thus the formation must adapt to the road or route shapes. Secondly, each vehicle as well as the entire convoy/fleet must avoid collisions with other vehicles, within or outside the formation. Thirdly, convoys/fleets must be .flexible so that mission planning can be recon.figured if necessary.

Key points of this work item are:
  • Usage of cooperative aerial imaging data for flight formation (over best routes) and mission planning (e.g. search and rescue).
  • xploiting fog computing (including aerial and ground elements) for timely feedback towards mission planning strategies.
C) Reliable Communications and Fast Computation
Communication technologies allow autonomous vehicles to share sensing information, control information (e.g. turning angle, speed) and the states of detected objects, allowing autonomous vehicles to perceive beyond their line-of-sight. Shared data among autonomous vehicles can improve system performance, since the view point of each vehicle can be very different, and thereby improve awareness. For instance, sharing planned trajectories may avoid potential motion conflicts, by means of motion coordination algorithms that can get to a consensus about solutions that are jointly feasible.

However, the computational requirements of autonomous vehicles are very large if we consider that video data may generate 20-40 MB/sec, radar around 10-100 KB/sec, and Lidar around 10-70 MB/sec (by Intel data). Hence, the workload of autonomous vehicles will require a new advanced network architecture that will encompasses not only inter-vehicle communication but also 5G networking and fog computing.

The envisioned large number of devices brings more complexity to network configuration and adaptation.  In this aspect, the concept of Software Defined Networking (SDN) may be applied for the management of network services through abstraction of low-level functionality to enable a programmable behavior suitable for the envisioned dynamic deployment scenarios. In the context of the CRAVE project the data plane is responsible to know what to do when new data is gathered locally or from intermittently available neighbor devices. This goes beyond current research on SDN, which targets packets forwarding on fixed infrastructure-based networks.

The functionality encompassed in the data plane is dynamically deployed in local virtual machines, allowing different virtual networks to co-exist in the same physical infrastructure. This separation of network services from devices follows the concept of Network Function Virtualization (NFV) and services are deployed based on a fog computing framework.

The fog network typically takes the form of nodes arranged in a hierarchy between the low-level control computers in the vehicle and the remote servers in the cloud. This hierarchical architecture aims to allow seamless, real-time communications and computation, which may not be achievable by cloud transactions alone.

Application of SDN/NFV/FOG paradigms helps to overcome the limitation of autonomous vehicle systems developed for specific applications.

Key points of this work item are:
  • Coordination of distributed fog devices for low latency computations.
  • Improve reliable communications over fading wireless channels, and environments with frequent changes in the network topology.
  • Cooperative relaying and opportunistic routing between vehicles to overcome intermittent communication infrastructure.

Research Team
  • Paulo Mendes, (PI)
  • Asaamining Godwin, NEMPS PhD Student.
  • Hector Orrillo, NEMPS PhD Student.

Experimental Environment

The goal is to evaluate the performance of real applications over a virtual network. For that we aim to use an emulator to connect real devises to a network simulation environment.

Experiments will be performed based on:

  • NETSIM emulator to evaluate wireless communications to support low latency computation.
  • Mininet-wifi emulator to evaluate the orchestration of software defined fog devices.

While Mininet-WiFi adds virtual wireless stations and access points to the Mininet emulator, the NETSIM emulator brings all the benefits of the NETSIM simulator namely:

  • Network modeling and planning:
    • Mobility models (Random waypoint, group mobility)
    • Path loss models (Indoor, free space, log distance)
    • Fading models (Rayleigh, Nakagami)
  • Traffic generator (database app, ftp, video, voice)
  • Optimize protocol performance
  • Good support for VANETS (802.11p, AODV).

Scientific Background

Cooperative Relaying among Mobile Devices (Selection of publications)

Tauseef Jamal, Paulo Mendes, “Cooperative Relaying in Dynamic Wireless Networks under Interference Conditions, IEEE Communications Magazine, Special issue on User-centric Networking and Services, December 2014.

Tauseef Jamal, Paulo Mendes, Andre Zuquete, "Wireless Cooperative Relaying Based on Opportunistic Relay Selection", International Journal On Advances in Networks and Services, Vol. 5, no. 1&2, July 2012.

Code for OMNET++ simulator

RelaySpot (2012): A Cooperative MAC Protocol for Dynamic Wireless Networks for Linux. OMNET++ module. (Authors: Tauseef Jamal, Paulo Mendes). Developed in the ULOOP project. SITI-SW-12-04

(more information about cooperative wireless relaying)


Opportunistic Wireless Routing (Selection of publications)

 Miguel Tavares, Omar Aponte, Paulo Mendes, "Named-data Emergency Network Services", in ACM MOBISYS, Munich, Germany, June 2018.

Christos-Alexandros Sarros, Sotiris Diamantopoulos, Sergi Rene, Ioannis Psaras, Adisorn Lertsinsrubtavee, Carlos Molina-Jimenez, Paulo Mendes, Rute Sofia, Arjuna Sathiaseelan, George Pavlou, Jon Crowcroft, Vassilis Tsaoussidis, "Connecting the Edges: A Universal, Mobile centric and Opportunistic Communications Architecture", IEEE Communication Magazine, February 2018.

P. Mendes, R. Sofia, V. Tsaoussidis, S. Diamantopoulos, C. Sarros, "Information-centric Routing for Opportunistic Wireless Networks", IETF Internet Draft (draft-mendes-icnrg-dabber-00), Feb 2018.

Code for Android

NDN-OPP (2017): Named-Data Networking (NDN) Android software package for Opportunistic Networks including multi-hop opportunistic wireless routing. Developed in the UMOBILE project. Available on Github.

(more information on Information centrric wireless networking)