|
Introduction to the workshop 8:25
Session I: Localization & mapping 8:30
Chairman: C. Laugier (INRIA, France)
-
Title: Navigable Maps for Intelligent Vehicles Localization and Perception 8:30
Keynote speaker: Philippe Bonnifait
(UTC, Compiegne, France)
35min + 5min questions
Presentation
Abstract:
Intelligent Vehicles are robotic systems that assist the driver in safe and comfortable operation by providing pertinent information or by controlling the vehicle itself. Real-time and safe perception of the driving environment is one of the key issues. Recent evolutions of navigable maps make them suitable to assist localization and perception processes since they provide additional information that can be exploited with anticipation.
This talk focuses on some autonomous techniques that merge the map information with on-board sensors data like GPS measurements, CAN -bus proprioceptive sensors, exteroceptive cameras and multi-layers lidars.
Macro-scale maps with poly-lines representation of the road network can be exploited as an a priori knowledge in order to enhance GPS availability, particularly in urban canyons where satellites signals are often blocked. Such kind of map technology is also planned to be used for Map-aided ADAS (Advanced Driver Assistance Systems). However, maps can be obsolete or contain errors, resulting in malfunctions of context-based ADAS and possibly generating hazardous situations. The talk will present a sequential fault detection test able to detect and localise map errors in an autonomous manner using the on-board sensors.
Meso-scale maps provide more refined information that describes the drivable space of the roads. The talk will present how 3-D facets geometry can be used for contracting East, North and altitude estimates when solving a localization problem. The use of this kind of 3D representation to characterize the drivable space (useful for path planning or obstacle avoidance) will be also presented and discussed.
Finally, the talk will focus on visual landmarks that can be managed in a specific layer of the map. A method for mobile mapping lane markings and exploiting them in dynamic localization will be described.
Experimental results showing the key role of navigable maps for intelligent vehicles localization and perception will be systematically presented.
- Title: Robot Localization using efficient planar features matching 9:10
Authors: B. Charmette, E. Royer, Frédéric Chausse and L. Lequievre
17min + 3min questions
Paper,
Presentation,
Video1
Abstract: Real-time accurate localization is a key component
of an autonomous mobile robot. Visual localization algorithms
usually rely on feature matching between the current view and
a map using point descriptors. Many descriptors such as SIFT
or SURF are designed to recognize features seen from different
viewpoint. But in a robotic context, the robot movement can
be modeled and bring useful information for the matching
problem. In this paper we detail a way of matching features with
a local 3D model of the features taking advantage of the motion
model of the robot. We describe then methods to describe the
motion model. The experimental results show how useful the
motion model of robot movement is, and prove that use of
other sensors can greatly improve precision and robustness of
the localization.
- Title: Application of Visual-Inertial SLAM for 3D Mapping of Underground Environments 9:30
Authors: A. Ferreira, J. Almeida and E. Silva
17min + 3min questions
Paper,
Presentation
Abstract:The underground scenarios are one of the most
challenging environments for accurate and precise 3d mapping
where hostile conditions like absence of Global Positioning
Systems, extreme lighting variations and geometrically smooth
surfaces may be expected. So far, the state-of-the-art methods
in underground modelling remain restricted to environments
in which pronounced geometric features are abundant. This
limitation is a consequence of the scan matching algorithms used
to solve the localization and registration problems.
This paper contributes to the expansion of the modelling
capabilities to structures characterized by uniform geometry and
smooth surfaces, as is the case of road and train tunnels. To
achieve that, we combine some state of the art techniques from
mobile robotics, and propose a method for 6DOF platform positioning
in such scenarios, that is latter used for the environment
modelling.
A visual monocular Simultaneous Localization and Mapping
(MonoSLAM) approach based on the Extended Kalman Filter
(EKF), complemented by the introduction of inertial measurements
in the prediction step, allows our system to localize himself
over long distances, using exclusively sensors carried on board a
mobile platform. By feeding the Extended Kalman Filter with
inertial data we were able to overcome the major problem
related with MonoSLAM implementations, known as scale factor
ambiguity. Despite extreme lighting variations, reliable visual
features were extracted through the SIFT algorithm, and inserted
directly in the EKF mechanism according to the Inverse
Depth Parametrization. Through the 1-Point RANSAC (Random
Sample Consensus) wrong frame-to-frame feature matches were
rejected.
The developed method was tested based on a dataset acquired
inside a road tunnel and the navigation results compared with a
ground truth obtained by post-processing a high grade Inertial
Navigation System and L1/L2 RTK-GPS measurements acquired
outside the tunnel. Results from the localization strategy are
presented and analyzed.
Session II: Multiple Vehicles/Robots & Interaction 9:50
Chairman: P. Bonnifait (Heudiasyc, France)
-
Title: Thoughts on perception for intelligent vehicles 9:50
Keynote speaker: Alberto Broggi (Parma University, Parma, Italy)
35min + 5min questions
Co-Authors: P. Grisleri and P. Zani
Presentation
Abstract: Many successful implementations of intelligent vehicles are
using laser based technology to obtain a 360 degrees view of the
area surrounding the vehicle. Such technology is able to provide
very dense 3D point clouds covering an extended range. The talk
will compare this technology to other lower cost alternatives,
such as vision, and discuss some possible implementations.
Coffee Break 10:30 ==> 11:00
- Title: Multiple Robots in a Cooperating Task: Exploration and Mapping 11:00
Authors: A.M. Neto, P. Rosa, T.E. Alves de Oliveira and P.C. Pellanda
17min + 3min questions
Paper,
Presentation
Abstract: This work presents an exploration task with multiple
vehicles using occupancy grids and a technique of simultaneous
localization and mapping (SLAM). The exploration strategy
uses concepts of costs and utility from frontier-cells. Besides,
the used SLAM method is based on a FastSLAM algorithm
with landmarks extracted from visual sensors and a features
map common to two vehicles. Both activities - location of the
vehicles and exploration of the environment - are coordinated
by a central agent. The results show that when two vehicles
can communicate with a central agent building a features map
common to the vehicles, the exploration task becomes more
efficient than when performed with dedicated maps, because
the accuracy of vehicle position and orientation is increased
with the use of an even number of particles. In this paper we
also present and evaluate the implementation of the approach
in a real environment.
- Title: Dynamic Obstacle Avoidance Strategies using Limit Cycle for the Navigation of Multi-Robot System 11:20
Authors: A. Benzerrouk, L. Adouane and P. Martinet
17min + 3min questions
Paper,
Presentation,
Video1,
Video2
Abstract: This paper deals with the navigation of a multirobot
system (MRS). The latter must reach and maintain a
specific formation in dynamic environment. In such areas,
the collision avoidance between the robots themselves and
with other obstacles (static and dynamic) is a challenging
issue. To deal with it, a reactive and a distributed control
architecture is proposed. The navigation in formation of the
MRS is insured while tracking a global virtual structure. In
addition, according to the robots’ perception context (e.g., static
or dynamic obstacle), the most suitable obstacle avoidance
strategy is activated. These approaches use mainly the limitcycle
principle and a penalty function to obtain linear and
angular robots’ velocities. The proposed control law guarantees
the stability (using Lyapunov function) and the safety of
the MRS. The robustness and the efficiency of the proposed
control architecture is demonstrated through a multitude of
experiments which shows the MRS in different configuration
of avoidance.
Lunch break 12:00
Session III: Interactive session 13:30
Chairman: P. Martinet (IRCCYN/ECN, Nantes, France)
- Title: Development of an Autonomous Vehicle for High-speed Navigation and Obstacle Avoidance
Authors: J.H. Ryu, D. Ogay, S. Bulavintsev, H. Kim, and J.S.Park
Paper
Abstract: This paper introduces the autonomous vehicle
Pharos, which participated in the 2010 Autonomous Vehicle
Competition organized by Hyundai-Kia motors. Pharos was
developed for high-speed on/off-road unmanned driving avoid-
ing diverse patterns of obstacles. For the high speed traveling
up to 60 Km/h, long range terrain perception, real-time path
planning and high speed vehicle motion control algorithms
are developed. This paper describes the major hardware and
software components of our vehicle.
- Title: Kinodynamic motion planning with state Lattice Motion Primitives
Authors: M. Pivoraiko and A. Kelly
Paper
Abstract: This paper presents a type of motion primitives than can be used for building efficient kinodynamic motion planners. The primitives are pre-computed to meet two obectives: to capture the mobility constraints of the robot as well as possible and to establish a state sampling policy that is conducive to efficent search. The first objective allows encoding mobility constraints into primitives, thereby enabling fast unconstrained search to produce feasible solutions. The second objective enables high quality (lattice) sampling of state space, further speeding up exploration during search. We further discuss several novel results enabled by using such motion primitives for kinodynamic planning, including incremental search, efficient bi-directional search and incremental sampling.
- Title: Detection of Moving and Stationary Objects at High Velocities using Cost-Efficient Sensors, Curve-Fitting and Neural Networks
Authors: F. Mirus, J. Pfadt, C. Connette, B. Ewert, D. Grudl, A. Verl
Paper,
Poster
Abstract: In recent years, driver-assistance systems have
emerged as one major possibility to increase comfort and –
even more important – safety in road traffic. Still, cost is one
major hindrance to the widespread use of safety systems such as
lane change or blind spot warning. To facilitate the widespread
adoption of such assistance systems, thus increasing safety for
all traffic participants, the use of cost-efficient components is
of crucial importance.
This paper investigates the usage of cost-efficient, widely used
ultrasonic sensors for blind spot warning at high velocities.
After discussing the requirements and setup of such a system
a model-based approach for the detection of moving and stationary
objects is outlined. The sensor-signal is compared with
a precalculated curve data base and the correlation-coefficients
are feeded into a neural network. To revise its performance the
concept at hand is qualitatively and quantitatively evaluated in
real road traffic situations under different driving conditions.
- Title: ESTRO: Design and Development of Intelligent Autonomous Vehicle for Shuttle Service in the ETRI
Authors: J. Byun, K.I. Na, M. Noh and S. Kim
Paper,
Poster
Abstract:
ESTRO(ETRI Smart Transport RObot) Project aims at the development of autonomous vehicle to transport goods and people without the help of driver within the well-structured section such campus and premises without traffic regulations. To do so, we have designed and implemented the autonomous vehicle modified electronic vehicle. In addition to we have constraints that have to minimize the cost of sensor and optimize the complexity of system unlike autonomous vehicles introduced in recent years for driving in urban traffic scenarios. This paper proposes the design of H/W and S/W architecture for the autonomous vehicle and describes the method of environmental perception and navigation. The implemented system is currently has been operational test in our institute campus.
- Title: An effective 6DoF motion model for 3D-6DoF Monte Carlo Localization
Authors:
A. L. Ballardini, A. Furlan, A. Galbiati, M. Matteucci, F. Sacchi, D. G. Sorrenti
Paper
Abstract: This paper deals with the probabilistic 6DoF
motion model of a wheeled road vehicle. It allows to correctly
model the error introduced by dead reckoning. Furthermore, to
stress the importance of an appropriate motion model, i.e., that
different models are not equally good, we show that another
model, which was previously developed, does not allow a correct
representation of the uncertainty, therefore misguiding 3D-
6DoF Monte Carlo Localization. We also present experiments,
in simulated settings as well as on field, to demonstrate that
our model allow a consistent determination of the 6DoF vehicle
pose.
- Title: Visual trajectory learning and following in unknown routes for autonomous navigation
Authors: D. A. Marquez-Gamez and M. Devy
Paper
Abstract: This paper describes the design and testing of
a system to enable large scale cooperative navigation of
autonomous vehicles moving on a priori unknown routes.
A large-scale learning-mapping approach and a map-based
replay-localization method are combined to achieve cooperative
navigation. The mapping approach is based on a proposed hierarchical/
hybrid BiCam SLAM approach -global level and local
maps-, which will be generalized to be executed on multiple
vehicles moving as a convoy. A global 3D map maintains the
relationships between a series of local maps built by the first
vehicle of the convoy (leader), defining a path that all other
vehicles (followers) must stay on. Only single camera setups
are considered. The overall approach is evaluated with real
data acquired in an urban environment.
- Title: Eigen analysis and gray alignment for shadow detection applied to urban scene images
Authors: T. Souza, L. Schnitman and L. Oliveira
Paper
Abstract: Urban scene analysis is very useful for many
intelligent transportation systems (ITS), such as license plate
detection, pedestrian detection, video surveillance, and so forth.
In fact, all these systems are prone to noisy of any kind, which
harms their performance. Among noisy factors, shadow may
represent a critical line between the success or fail within an
ITS framework. Shadow detection usually provides benefits for
further stages of machine vision systems, and its application
will depend on the computational load of the detection system.
To cope with all these situations, a novel shadow detection
method applied to urban scenes is proposed here. This method
is based on a measure of the energy defined by the summation
of the eigenvalues of image patches. The final decision of an
image region to contain a shadow is made according to a new
metric for unsupervised classification called here as a gray
alignment. The characteristics of the proposed method include
no supervision, very low computational cost and mathematical
background unification, which turns the method effective.
Our proposed approach was evaluated on a public dataset,
demonstrating state-of-the-art performance.
Session IV: Navigation, Control, Planning 14:30
Chairman: P. Grisleri (Parma University, Parma, Italy)
-
Title: Interpretation of Situative Sensor-Data and Continuous Decision Making for Cognitive Automobiles 14:30
Keynote speaker:Rüdiger Dillmann (KIT, Karlsruhe, Germany)
35min + 5min questions
Co-Authors: S. Brechtel, T. Gindele
Presentation
Abstract: Autonomous automobiles must be capable to decide continuously on their adequate behaviour in a highly dynamic environment.
Usually such vehicles are supported with uncomplete and noisy perceptive data with different sensor modalities. Probabilistic and predictive methods can be applied to estimate the
actual situationand the intension of other traffic agents around the autonomous vehicle and to generate the best
and safe behaviour. Predictive driving requires understanding the sensorial observations and the behaviour of the other vehicles in standard and non standard
traffic situations. With such a background, structural bootstrapping towards not befor seen situations can be realized and logically explained with the help of learning strategies.
A probate method is to collect automatically experience from traing data and online observations in form of Bayesian Nets. For
processing of the consequences of actions under uncertainties continuous POMDPs may be suitable to generate adequate decisions.
Planning algorithms, decision problems and its implementation and experimental verification as well as actual research results are presented in this talk.
- Title: An Efficient Heuristic Estimate for Non-holonomic Motion Planning 15:10
Authors: J.W. Choi
17min + 3min questions
Paper,
Presentation
Abstract: A new efficient and admissible heuristic estimate
function is proposed for non-holonomic motion planning. The
heuristic calculation begins by separating relatively open local
area around the goal from others. Then pre-computed heuristics
in obstacle-free full state space are assigned to the area. The
heuristic for the other area is obtained by using dynamic
programming to extende the full state heuristic in reduced 2D
state space. The numerical simulations demonstrate remarkable
performance improvement by applying the heuristic function,
compared to other existing heuristics.
- Title: Short term path planning using a multiple hypothesis evaluation approach for an autonomous driving competition 15:30
Authors: M. Oliveira, V. Santos and A.D. Sappa
17min + 3min questions
Paper,
Presentation,
Video1,
Video2,
Video3,
Video4
Abstract: This paper describes a practical implementation
of short term path planning in autonomous navigation in
unmapped or unstructured environments. Path planning is
performed by generating multiple hypothesis of paths for the
robot and then evaluating the quality of each path. In very
dynamic environments, long term path planning is generally
not very useful, so this paper embraces the approach of short
term path planning and continuously revises the path plan
and motion parameters after the perception from its onboard
sensors. The solution has been applied to small scale robots that
compete in an autonomous driving competition. These robots
have won the last six editions of this competition.
Coffee break 16:00
Session V: Perception & Situation awareness 16:30
Chairman: C. Laugier (INRIA, Grenoble, France)
- Title: Real-time Scan-Matching Using L0-norm Minimization Under Dynamic Crowded Environment 16:30
Authors: Y. Hieida, T. Suenaga, K. Takemura, J. Takamatsu and T. Ogasawara
17min + 3min questions
Paper,
Presentation
Abstract: We propose real-time scan-matching based on L0-
norm minimization under dynamic crowded environment. The
prior scan-matching methods are based on L2-norm minimization,
because the measurement noise follows the normal distribution
in static environments. This assumption is unfortunately
broken in dynamic crowded environments.
We propose to use the idea of Locality Sensitive Hashing
(LSH) to accelerate the L0-norm minimization, which usually
is a time-consuming process. The LSH customized for our issue
reduces the calculation time even in the worst cases. The experimental
results demonstrate the effectiveness of the proposed
method compared with standard L2-norm minimization and its
robust version with M-estimator.
- Title: Fast classification of static and dynamic environment for Bayesian Occupancy Filter (BOF) 16:50
Authors: Q. Baig, M. Perrollaz, J. Botelho Do Nascimento, C. Laugier
17min + 3min questions
Paper,
Presentation,
Video1
Abstract: In this paper we present a fast motion detection
technique based on laser data and odometry/imu information.
This technique instead of performing a complete SLAM (Simultaneous
Localization and Mapping) solution, is based on
transferring occupancy information between two consecutive data
grids. We plan to use the output of this work for Bayesian
Occupancy Filter (BOF) framework to reduce processing time
and improve the results of subsequent clustering and tracking
algorithm, based on BOF. Experimental results obtained from a
real demonstrator vehicle show the effectiveness of our technique.
-
Title: Localization and Mapping in Dynamic and Changing Environments 17:10
Keynote speaker: Wolfram Burgard (Freiburg University, Frieburg, Germany)
35min + 5min questions
Co-Authors: Gian Diego Tipaldi
Paper,
Presentation,
Video1,
Video2,
Video3,
Video4,
Video5,
Video6,
Video7
Abstract: The majority of existing approaches to mobile robot localization and mapping assumes that the world is static, ignoring the dynamics inherent in most real world scenarios like parking lots, warehouses and even offices and households. In such environments the configuration of certain objects such as cars, goods, or furniture can change with time leading to inconsistent observations with respect to previously learned maps and thus decreasing the localization accuracy.
In this talk we present a probabilistic grid-based approach for modeling changing environments. Our method represents both, the occupancy and its changes in the corresponding area where the dynamics are characterized by the state transition probabilities of a Hidden Markov Model.
We further present a novel probabilistic approach to lifelong localization in changing environments, where the robot pose and the environment state are jointly estimated using a Rao-Blackwellized particle filter. Exploiting several characteristics of HMMs, we can considerably speed up the estimation procedure. This makes it feasible to run our algorithm on-line. Experimental results obtained with data acquired by real robots demonstrate that our model is well-suited for representing changing environments.
Further results demonstrate that our approach can reliably adapt to changes in the environment and that it significantly improves standard localization techniques.
Closing 17:50
|
