Chairman: Philippe Martinet
Session I: Perception & Localization 8:35-10:10
Chairman: Alberto Broggi
Session II: Path Planning & Navigation systems 10:30-12:05
- Title: The VIAC Challenge: Setup of an Autonomous Vehicle for a 13,000 km Intercontinental Unmanned Drive 8:35-9:20
Keynote speaker: Alberto Broggi (Parma University, Italy) 40min + 5min questions
Co-Authors: Massimo Bertozzi, Luca Bombini, Alberto Broggi, and Paolo Grisleri
Abstract: Autonomous vehicles have been demonstrated to be able to traverse the desert (the DARPA Grand Challenge, 2005),
navigate downtown together with other traffic (the DARPA Urban Challenge, 2007), someone is even trying to emulate experienced drivers in extreme races,... In all these situations, however,
the unmanned vehicles move within a semi-controlled environment. VisLab is now trying to push the unmanned vehicles technology to the limit and test their systems (both hardware and software) for a
long time and in an extreme environment: on July 10, 2010, two autonomous vehicles will leave Italy and will drive for 13,000 km in Europe towards Moscow, then Russia, then Siberia, Kazakstan, then
China, Mongolia, finally reaching Shanghai on October 10, 2010, after 3 months of autonomous driving. As a 'challenge into the challenge, VisLab selected electric vehicles, with the final aim of setting
a new milestone in the history of robotics: goods will be transported from Italy to China on a ground trip with no driver, and without using a drop of conventional fuel. Not only these vehicles will be
moving without any human intervention, but the driverless technology will be powered by solar energy thanks to a panel on the vehicle's roof. The talk will present the current state of the art and the major
- Title: Learning a Real-Time 3D Point Cloud Obstacle Discriminator via Bootstrapping 9:20-9:45
Authors: Michael Samples and Michael R. James 20min + 5min questions
Abstract: Many recent mobile robotics applications have
incorporated the use of 3D LIDAR sensors as an important
component in scene understanding due to frequent data measurements
and direct observation of geometric relationships.
However, the sparseness of point cloud information and the lack
of unique cues at an individual point level presents challenges
in algorithm design for obstacle detection, segmentation, and
tracking. Since individual measurements yield less information
about the presence of obstacles, many algorithmic approaches
model the joint posterior of point-labels. Such approaches can
produce robust point labelings at higher computation cost. In
this paper, we apply joint posterior approaches with smooth
terrain priors for point cloud obstacle discrimination. The
resulting labels are used to bootstrap efficient discriminators
which require no human labeled data, yet are comparable in
discriminative ability to the joint posterior approaches.
- Title: In Improved Flies Method for Stereo Vision: Application to Pedestrian Detection 9:45-10:10
Authors: Hao Li, Gwenaelle Toulminet, Fawzi Nashashibi 20min + 5min questions
Abstract: In the vast research field of intelligent transportation systems, the problem of detection (and recognition) of environment objects, for example
pedestrians and vehicles, is indispensable but challenging. The research work presented in this paper is devoted to stereo-vision based method with pedestrian detection as its application (a sub-part of the French national project
"LOVe": Logiciels d'Observation des Vulnerables). With a prospect of benefiting from an innovative method i.e. the genetic evolutionary "flies" method proposed by former researchers on continuous data updating and asynchronous
data reading, we have carried on the "flies" method through the task of pedestrian detection affiliated with the "LOVe" project. Compared with former work of the "flies" method, two main contributions have been incorporated into
the architecture of the "flies" method: first, an improved fitness function has been proposed instead of the original one; second, a technique coined "concentrating" has been integrated into the evolution procedure. The improved "flies"
method is used to offer range information of possible objects in the detection field. The integrate scheme of pedestrian detection is presented as well. Some experimental results are given for validating the performance improvements
brought by the improved "flies" method and for validating the pedestrian detection method based on the improved "flies" method.
Chairman: Rüdiger Dillman
Session III: Human Robot Interaction 13:30-14:15
- Title: Situation Assessment and Behaviour Decision Making of Cognitive Vehicles 10:30-11:15
Keynotes speaker: Rüdiger Dillman (Karlsruhe University, Germany) 40min + 5min questions
Abstract: Driving an autonomous vehicle on urban and rural environment requires knowledge about the situation on the road. Knowledge about the intension of other vehicles and inividuals on the road is required in order to classify the situation and to decide how to behave and to react. This paper addresses the problem of extracting information about the situative traffic environment of a vehicle from ist sensorial observations, its interpretation referencing situative knowledge and an estimation of itīs further behaviour. This estimation requires understanding the intension of the other vehicles or agents and a predictive view of further traffic state evolvement. Because of uncomplete observation and uncertainties the estimation and sensor fusion process has an important role. With the help of learning methods in terms of learning from example the vehicle will be able to learn from ist observations which allows the estimation of dangerous situations and a predictive view of its environment which allows the continuation of driving.
Furthermore it is necessary to make according the actual situation and drive intension behavioural decisions considering itīs effects and results. Also here uncertainty has to be considered to enable predictive driving. A predictive behavioural decision process in combination with a learning process will be presented which allows to enhance the decision performance. A dynamic risk map is used to support algorithms for motion planning of the vehicle.
Finally the vehicle should be able to execute maneuvers such as passing a crossing, lane changing, collision avoidance, overtaking and turning off, processing information about the actual situation and a prediction how it may evolve.
The work to be reported is part of the collaborative research center SFB/TR 28 Cognitive Automobile which is sponsored by the German Research Agency DFG.
- Title: Optimal Vehicle Routing and Scheduling with Precedence Constraints and Location Choice 11:15-11:40
Authors: G. Ayorkor Korsah, Anthony Stentz, and M. Bernardine Dias, and Imran Fanaswala 20min + 5min questions
Abstract: To realize the vision of intelligent transportation
systems with fully automated vehicles, there is a need for highlevel
planning for single vehicles as well as fleets of vehicles.
This paper addresses the problem of optimally assigning and
scheduling a set of spatially distributed tasks to a fleet of vehicles
working together to achieve a high-level goal, in domains
where tasks may be related by precedence or synchronization
constraints and might have a choice of locations at which they
can be performed. Such problems may arise, for example, in
disaster preparedness planning, transportation of people, and
delivery of supplies. We present a novel mathematical model of
the problem and describe how it can be solved optimally in a
- Title: Multi-Agent Planning and Simulation for Intelligent Transportation System 11:40-12:05
Authors: Ming C. Lin, Jason Sewall, Jur Van den Berg, David Willkie, Dinesh Manocha 20min + 5min questions
Abstract: In this paper, we provide a brief survey of our recent
work on multi-agent planning and simulation for intelligent
transportation system. In particular, we first present a novel
algorithm to reconstruct and visualize continuous traffic flows
from discrete spatio-temporal data provided by traffic sensors.
Given the positions of each car at two recorded locations on a
highway and the corresponding time instances, our approach
can reconstruct the traffic flows (i.e. the dynamic motions of
multiple cars over time) in between the two locations along the
highway using a priority-based multi-agent planning algorithm.
Our algorithm is applicable to high-density traffic on highways
with an arbitrary number of lanes and takes into account the
geometric, kinematic, and dynamic constraints on the cars. In
addition, we describe an efficient method for simulating realistic
traffic flows on large-scale road networks. Our technique is
based on a continuum PDE model of traffic flow that we
extend to correctly handle lane changes and merges, as well as
traffic behaviors due to changes in speed limit. We show that
our method can simulate plausible traffic flows on publiclyavailable,
real-world road data and demonstrate the scalability
of this technique on many-core systems.
Session IV: Interactive session 14:15-15:45
- Title: Robot for the Human 13:30-14:15
Keynote speaker: Oussama Khatib (Stanford University, USA) 40min + 5min questions
Abstract: Robotics is rapidly expanding into the human environment and vigorously engaged in its new emerging challenges.
From a largely dominant industrial focus, robotics has undergone by the turn of the new millennium a major transformation in scope and dimensions. This expansion has been brought about by the maturity
of the field and the advances in its related technologies. The new generation of robots is expected to safely and dependably co-habitat with humans in homes, workplaces, and communities, providing
support in services, entertainment, education, health care, manufacturing, and assistance. Interacting, exploring, and working with humans, the new generation of robots will increasingly touch people
and their lives. New design and fabrication concepts, novel sensing modalities, effective planning and control strategies, modeling and understanding of human motion and skills are among the key
requirements discussed for the development of this new generation of human-friendly robots.
Chairman: Philippe Martinet
Session V: Multi-Robot Control & ITS 15:45-17:20
- Title: Benchmark Tools for Evaluating AGVs at Industrial Environments
Authors: Hector Yuste, Leopoldo Armesto and Josep Tornero
Abstract: The paper addresses the problem of evaluating
AGVs with different degrees of autonomy by defining a methodology
and benchmark tools to grade the performance of each
solution. The proposed benchmark requires running different
experiments, from manual driving to autonomous navigation,
at different velocities and different scenarios. The goal is to
evaluate the performance of AGVs, in terms of robustness
to reach the goal, collisions reduction, traveling time, average
speed, etc. The underlying objective is to evaluate the potential
advantages of manual-assisted driving as well as autonomous
navigation against standard manual driving. To obtain valid
and significant results, 180 experiments have been completed
on each case with drivers of different ages, sex and skills.
- Title: Automatic Routing System for Intelligent Warehouses
Authors: K. T. Vivaldini, J. P. M. Galdames, T. B. Pasqual, R. M. Sobral; R. C. Araújo, M. Becker, and G. A. P. Caurin
Abstract: Automation of logistic processes is essential to
improve productivity and reduce costs. In this context,
intelligent warehouses are becoming a key to logistic systems
thanks to their ability of optimizing transportation tasks and,
consequently, reducing costs. This paper initially presents
briefly routing systems applied on intelligent warehouses. Then,
we present the approach used to develop our router system.
This router system is able to solve traffic jams and collisions,
generate conflict-free and optimized paths before sending the
final paths to the robotic forklifts. It also verifies the progress
of all tasks. When a problem occurs, the router system can
change the tasks priorities, routes, etc. in order to avoid new
conflicts. In the routing simulations each vehicle executes its
tasks starting from a predefined initial pose, moving to the
desired position. Our algorithm is based on Dijkstra's shortestpath
and the time window approaches and it was implemented
in C language. Computer simulation tests were used to validate
the algorithm efficiency under different working conditions.
Several simulations were carried out using the Player/Stage
Simulator to test the algorithms. Thanks to the simulations, we
could solve many faults and refine the algorithms before
embedding them in real robots.
- Title: Coordinating the motion of multiple AGVs in automatic warehouses
Authors: Roberto Olmi, Cristian Secchi and Cesare Fantuzzi
Abstract: In this paper an algorithm for planning a coordinated
motion of a fleet of Autonomous Guided Vehicles (AGVs)
delivering goods in an automatic warehouse is proposed. The
AGVs travel along a common segmented layout and a path
is assigned to each robot by a mission planner. Coordination
diagrams are used for representing possible collisions among
the robots and a novel algorithm for efficiently determining a
coordinated motion of the fleet is proposed. The coordination
approach proposed in the paper is validated through experiments
on real plants layouts. We present an example in which
the coordinated motion of 10 vehicles is computed in only 12.4
sec. on a common PC.
- Title: ArosDyn: Robust Analysis of Dynamic Scenes by means of Bayesian Fusion of Sensor Data - Application to the Safety of Car Driving
Authors: Christian Laugier, Igor E. Paromtchik, Mathias Perrollaz, Mao Yong, Amaury Nčgre, John-David Yoder, Christopher Tay
Abstract: The ArosDyn project aims to develop an embedded
software for robust analysis of dynamic scenes in urban
environment during car driving. The software is based on
Bayesian fusion of data from telemetric sensors (lidars) and
visual sensors (stereo camera). The key objective is to process
the dynamic scenes in real time to detect and track multiple
moving objects, in order to estimate and predict risks of
collision while driving.
- Title: Real-Time Detection of Moving Obstacles from Mobile Platforms
Authors: Chunrong Yuan and Hanspeter A. Mallot
Abstract: In this paper we present a vision-based algorithm
for the detection of moving obstacles in complex and unknown
environments. The goal is to find moving objects from images
captured by a mobile camera navigating together with a
moving platform. One specific feature of our approach is
that it does not need any information of the camera and
hence works without camera calibration. Another advantage
lies in the fact that it integrates motion separation and outlier
detection into one statistical framework. Based on sparse point
correspondences extracted from consecutive frame pairs, scene
points are clustered into different classes by statistical analysis
and modeling of the probability distribution function of the
underlying motion characteristics. Experimental results based
on several real-world video streams demonstrate the efficiency
of our algorithm.
- Title: Studying of WiFi range-only sensor and its application to localization and mapping systems
Authors: F. Herranz, M. Ocaņa, L. M. Bergasa, M. A. Sotelo, D. F. Llorca, N. Hernandez, A. Llamazares and C. Fernandez
Abstract: The goal of this paper is to study a noisy WiFi
range-only sensor and its application in the development of
localization and mapping systems. Moreover, the paper shows
several localization and mapping techniques to be compared.
These techniques have been applied successfully with other
technologies, like ultra-wide band (UWB), but we demonstrate
that even using a much more noisier sensor these systems can
be applied correctly. We use two trilateration techniques and a
particle filter to develop the localization and mapping systems
based on the range-only sensor. Some experimental results and
conclusions are presented.
- Title: Terrain Classification for Improving Traversability of Autonomous Vehicles
Authors: Jayoung Kim, Jonghwa Lee, Jihong Lee
Abstract: One of the requirements for autonomous vehicles
on off-road is to move harmoniously in unstructured
environments. It is an undeniable fact that such capacity of
autonomous vehicles is the most important in an aspect
considering mobility of the vehicle. So, many researchers use
contact and/or non-contact methods to detect a terrain whether
the vehicle can move on or not. In this paper we introduce an
algorithm to classify terrains using visual information. As
pre-processing, contrast enhancement technique is introduced
to improve accurate rate of classification. Also, for conducting
classification algorithm, training images are grouped as each
material and Bayesian classification recognizes new images as
each material using such material groups. Consequently, we
can confirm the good performance of classification. Moreover,
we can build Traversability map on which autonomous vehicles
can predict whether to go or not to go through real friction
coefficients which are measured by Load-Cell on surfaces of
Chairman: Philippe Martinet
- Title: Mobile Millenium 15:45-16:30
Keynote speaker: Alexandre Bayen (Berkeley University, USA) 40min + 5min questions
Abstract: This talk describes how the mobile internet is changing the face of traffic monitoring at a rapid pace. In the last five years, cellular phone technology has bypassed several attempts to construct dedicated infrastructure systems to monitor traffic. Today, GPS equipped smartphones are progressively morphing into an ubiquitous traffic monitoring system, with the potential to provide information almost everywhere in the transportation network. Traffic information systems of this type are one of the first instantiations of participatory sensing for large scale cyberphysical infrastructure systems.
However, while mobile device technology is very promising, fundamental challenges remain to be solved to use it to its full extent, in particular in the fields of modeling and data assimilation. The talk will present a new system, called Mobile Millennium, launched recently by UC Berkeley, Nokia and Navteq, in which the driving public in Northern California can freely download software into their GPS equiped smartphones, enabling them to view traffic in real time and become probe vehicles themselves.
The smartphone data is collected in a privacy-by-design environment, using spatially aware sampling. Using data assimilation, the probe data is fused with existing sensor data, to provide real time estimates of traffic.
Results from experimental deployments in California and New York will be presented, as well as preliminary results from a pilot field operational test in California, with already more than 5,000 downloads.
- Title: A global decentralized control strategy for urban vehicle platooning relying solely on monocular vision 16:30-16:55
Authors: Pierre Avanzini, Benoit Thuilot, Eric Royer, Philippe Martinet 20min + 5min questions
Abstract: Automated electric vehicles available in free access
constitute a promising very efficient and environment-friendly
"urban transportation system". An additional functionality that
could enhance this transportation service is vehicle platooning.
In order to avoid oscillations within the platoon when
completing this task, a global control strategy, supported by
inter-vehicle communications, is investigated. Vehicle absolute
localization is then needed and is here derived from monocular
vision. These data are however expressed in a virtual vision
world, slightly distorted with respect to the actual metric one.
It is shown that such a distortion can accurately be corrected
by designing a nonlinear observer relying on odometric data. A
global decentralized control strategy, relying on exact linearization
techniques, can then be designed to achieve accurate vehicle
platooning. Simulations and full-scale experiments demonstrate
the performance of the proposed approach.
- Title: Lyapunov Global Stability for a Reactive Mobile Robot Navigation in Presence of Obstacles 16:55-17:20
Authors: Ahmed Benzerrouk, Lounis Adouane, Philippe Martinet 20min + 5min questions
Abstract: This paper deals with the navigation of a mobile
robot in unknown environment. The robot has to reach a final
target while avoiding obstacles. It is proposed to break the task
complexity by dividing it into a set of basic tasks: Attraction to a
target and obstacle avoidance. Each basic task is accomplished
through the corresponding elementary controller. The activation
of one controller for another is done according to the priority
task. To ensure the overall stability of the control system,
especially at the switch moments, properties of hybrid systems
are used. Hybrid systems allow switching between continuous
states in presence of discrete events. In this paper, it is proposed
to act on the gain of the proposed control law. The aim is to
ensure the convergence of a common Lyapunov function to all
the controllers. This ensures the stability of the overall control.
Simulation results confirm the theoretical study.
Chairman: Philippe Martinet