Session I: Motion planning 9:00-10:30
Chairman: Alonzo Kelly and Christian Laugier
- Title: Fast and Feasible Deliberative Motion Planner for Dynamic Environments (invited paper)
Authors: Mihail Pivtoraiko and Alonzo Kelly
Paper,
Presentation
Abstract
We present an approach to the problem of differentially
constrained mobile robot motion planning in arbitrary
time-varying cost fields. We construct a special search space
which is ideally suited to the requirements of dynamic environments
including a) feasible motion plans that satisfy differential
constraints, b) efficient plan repair at high update rates, and
c) deliberative goal-directed behavior on scales well beyond
the effective range of perception sensors. The search space
contains edges which adapt to the state sampling resolution yet
aquire states exactly in order to permit the use of the dynamic
programming principle without introducing infeasibility. It is a
symmetric lattice based on a repeating unit of controls which
permits off-line computation of the planner heuristic, motion
simulation, and the swept volumes associated with each motion.
For added planning efficiency, the search space features fine
resolution near the vehicle and reduced resolution far away.
Furthermore, its topology is updated in real-time as the vehicle
moves in such a way that the underlying motion planner
processes changing topology as an equivalent change in the
dynamic environment. The planner was originally developed
to cope with the reduced computation available on the Mars
rovers. Experimental results with research prototype rovers
demonstrate that the planner allows us to exploit the entire
envelope of vehicle maneuverability in rough terrain, while
featuring real-time performance.
- Title: Benchmarking Collision Avoidance Schemes for Dynamic Environments
Authors: Luis Martinez-Gomez and Thierry Fraichard
Paper,
Presentation,
video1
Abstract
This paper evaluates and compare three state-of-the-art collision avoidance schemes designed to operate in dynamic environments.
The first one is an extension of the popular Dynamic Window approach; it is henceforth called TVDW which stands for Time-Varying
Dynamic Window. The second one called NLVO builds upon the concept of Non Linear Velocity Obstacle which is a generalization of
the Velocity Obstacle concept. The last one is called ICS-Avoid, it draws upon the concept of Inevitable Collision States,
i.e. states for which, no matter what the future trajectory of the robotic system is, a collision eventually occurs.
The results obtained show that, when provided with the same amount of information about the future evolution of the environment,
ICS-Avoid outperforms the other two schemes. The primary reason for this has to do with the extent to which each collision
avoidance scheme reasons about the future. The second reason has to do with the ability of each collision avoidance scheme
to find a safe control if one exists. ICS-Avoid is the only one which is complete in this respect thanks to the concept of Safe Control Kernel.
- Title: Mapping Obstacles to Collision States for On-line Motion Planning in Dynamic Environments
Authors: Oren Gal and Zvi Shiller
Paper,
Presentation,
video1,
video2,
video3,
video4,
video5
Abstract
This paper presents a mapping of static and
moving obstacles using, Velocity Obstacles (VO), for on-line
planning in dynamic environments. Each obstacle is mapped
to forbidden states by selecting a proper time horizon for the
velocity obstacle. The proper choice of the time horizon ensures
that the boundary of the mapped obstacle overlaps with the
boundary of the set of inevitable collision states (ICS). This
time horizon is determined by the minimum time it would
take the robot to avoid collision, either by stopping or by
passing the respective obstacle. This mapping allows safe online
planning using only one step look ahead. The on-line
trajectories favorably compare with the trajectories obtained
by a global planner.
- Title: Probabilistic Rapidly-exploring Random Trees for autonomous navigation among moving pedestrians
Authors: Chiara Fulgenzi, Anne Spalanzani, and Christian Laugier
Paper,
Presentation
Abstract
The paper presents a navigation algorithm for
dynamic, uncertain environment. The static environment is
unknown, while moving pedestrians are detected and tracked
on-line. The planning algorithm is based on an extension
of the Rapidly-exploring Random Tree algorithm, where the
likelihood of the obstacles trajectory and the probability of
collision is explicitly taken into account. The algorithm is used
in a partial motion planner, and the probability of collision is
updated in real-time according to the most recent estimation.
Results show the performance for a car-like robot among a
pedestrian tracking dataset and simulated navigation among
multiple dynamic obstacles.
Session II: Multi-sensor perception & navigation 10:50-12:50
Chairman: Anna Petrovskaya and Martin Rufli
- Title: Multi-Sensor Perception and Dynamic Path Planning in City Environments (invited paper)
Authors: Martin Rufli Luciano Spinello Roland Siegwart
Paper,
Presentation
Abstract
In this paper we describe a state lattice based
path planning approach, which we have successfully applied to
large, cluttered, but quasi-static environments. Our approach
produces smooth and complex maneuvers through the use of
a multi-resolution state lattice, where the resolution is adapted
based on the environment, and distance from the robot.
We also describe a framework for detecting dynamic obstacles
such as pedestrians and cars using a multisensor lasercamera
detection and tracking method. Image detection is
based on several extensions to the Implicit Shape Model
technique; laser detection is instead achieved through the use of
a Conditional Random Fields reasoning. Objects are tracked
through the use of multiple motion model Kalman filters in
order to cope with several different motion dynamics.
Urban environments, are complex, cluttered, and dynamic
scenes, however. We therefore propose to extend our dynamic
obstacle detection and tracking method with a short-term
motion prediction functionality based on the same models used
for tracking, effectively generating time based cost or risk
maps. We further propose to implement these cost maps into
our high-dimensional (5D to 6D) lattice planner to generate
time-optimal trajectories in dynamic, cluttered environments.
A D* implementation is envisioned to speed up re-planning
dramatically.
- Title: Camera and Laser Radar Co-detection of Pedestrians
Authors: Hao LI, Ming YANG, Huijia QIAN
Paper,
Presentation
Abstract
Intelligent vehicle technology is a
promising technology for enhancing urban traffic safety and
efficiency. Pedestrian detection is an important issue for
applications of intelligent vehicles in urban environments.
The kind of most widely used method for pedestrian detection
is vision based method. The general problems for vision
based method are how to obtain a proper ROI (region of
interests) efficiently and how to detect and segment contours
of candidate objects out of ROI. In this paper, a camera and
laser radar co-detection method is proposed. First, a method
of camera and laser radar co-calibration is presented. Second,
a method of how to obtain proper ROI and the contours of
candidate objects using the co-calibration results is
introduced. Finally, a decision rule is induced from a set of
examples of contour shapes of both pedestrians and
landmarks (They are most likely to be confused with each
other because of their similarity in size). Some experimental
results are given for validating the camera and laser radar
co-detection method.
- Title: Model Based Vehicle Tracking in Urban Environments (invited paper)
Authors: Anna Petrovskaya and Sebastian Thrun
Paper,
Presentation,
video1,
video2,
video3,
video4
Abstract
Situational awareness is crucial for autonomous
driving in urban environments. We present the moving vehicle
tracking module we developed for our autonomous driving
robot Junior. The robot won second place in the Urban Grand
Challenge, an autonomous driving race organized by the U.S.
Government in 2007. The module provides reliable detection
and tracking of moving vehicles from a high-speed moving
platform using laser range finders. Our approach models both
dynamic and geometric properties of the tracked vehicles and
estimates them using a single Bayes filter per vehicle. We
show how to build consistent and efficient 2D representations
out of 3D range data and how to detect poorly visible black
vehicles. Experimental validation includes the most challenging
conditions presented at the Urban Grand Challenge as well as
other urban settings.
- Title: Connexity based fronto-parallel plane detection for stereovision obstacle segmentation
Authors: Thomas Veit
Paper,
Presentation,
video1,
video2,
video3,
video4,
video5,
video6,
video7,
video8
Abstract
Progress in hardware makes it possi-
ble to compute dense disparity maps in real-time.
This work describes a suitable obstacle segmentation
method for these dense disparity maps. The method
analyses the connexity of the disparity map in order
to extract fronto-parallel planes by means of a suit-
able depth constraint. This pragmatic geometrical ap-
proach reduces the number of detection parameters.
As a consequence it is easy and intuitive to use by
a non-expert end-user. The target application eld is
Advanced Driving Assistance Systems (ADAS). The
performance of the method is illustrated by various
results on real image sequences in the context of
pedestrian detection.
- Title: Safe and Dependable Operation of a Large Industrial Autonomous Forklift
Authors: Ashley Tews
Paper,
Presentation,
video1,
video2,
video3
Abstract
For autonomous vehicles to operate in industrial
environments, they must demonstrate safe, reliable, predictable,
efficient and repeatable performance. To achieve this, two
important high level factors are situational awareness and
system dependability. The vehicle must be able to identify
objects and predict the trajectories of dynamic objects in order
to avoid unplanned interaction and to improve performance.
In many environments, the vehicle is also required to operate
for long periods of time over many days, weeks and months.
Towards this goal, the vehicle needs to self-monitor its hardware
and software systems, and have redundant primary systems.
We have incorporated many of these requirements into our
Autonomous Hot Metal Carrier which is a modified 20 tonne
forklift used in aluminium smelters for carrying a 10 tonne
payload between large sheds, in the presence of other vehicles
and people. Our HMC has successfully conducted 100 of hours
of autonomous operation in our industrial worksite. The main
hardware and software systems will be discussed in this paper
with particular focus on the redundant localisation and obstacle
avoidance systems. Experiments are described to highlight the
performance of the HMC systems in the presence of dynamic
objects around a typical worksite.
Session III: Vision based perception & Visual SLAM 14:00-15:30
Chairman: François Chaumette and Philippe Martinet
- Title: Comparing appearance-based controllers for nonholonomic navigation from a visual memory (invited paper)
Authors: Andrea Cherubini, Manuel Colafrancesco, Giuseppe Oriolo, Luigi Freda and François Chaumette
Paper,
Presentation,
video1
Abstract
In recent research, autonomous vehicle navigation
has been often done by processing visual information. This
approach is useful in urban environments, where tall buildings
can disturb satellite receiving and GPS localization, while
offering numerous and useful visual features. Our vehicle uses
a monocular camera, and the path is represented as a series
of reference images. Since the robot is equipped with only
one camera, it is difficult to guarantee vehicle pose accuracy
during navigation. The main contribution of this article is the
evaluation and comparison (both in the image and in the 3D
pose state space) of six appearance-based controllers (one posebased
controller, and five image-based) for replaying the reference
path. Experimental results, in a simulated environment,
as well as on a real robot, are presented. The experiments
show that the two image jacobian controllers, that exploit the
epipolar geometry to estimate feature depth, outperform the
four other controllers, both in the pose and in the image space.
We also show that image jacobian controllers, that use uniform
feature depths, prove to be effective alternatives, whenever
sensor calibration or depth estimation are inaccurate.
- Title: A generic framework for topological navigation of urban vehicle
Authors: Jonathan Courbon, Youcef Mezouar, Laurent Eck, Philippe Martinet
Paper,
Presentation,
video1,
video2,
video3,
video4
Abstract
In this paper, we present a generic framework for
urban vehicle navigation using a topological map. This map is
built by taking into account the non-holonomic behaviour of the
vehicle. After a localization step, a sensory route is extracted to
reach a goal. This route is followed using a sensor-based control
strategy, based on the vehicle model and computed from the
state extracted from the current and the desired sensory images.
In that aim, a generic model is proposed for visual sensors.
Experiments with an urban electric vehicle navigating in an
outdoor environment have been carried out with a fisheye camera
using a single camera and natural landmarks. A navigation along
a 1700-meter-long trajectory validates our approach.
- Title: Use a Single Camera for Simultaneous Localization And Mapping with Mobile Object Tracking in dynamic environments
Authors: Davide Migliore, Roberto Rigamonti, Daniele Marzorati, Matteo Matteucci, Domenico G. Sorrenti
Paper,
Presentation
Abstract
The aim of this work is to demonstrate that it is
possible to use a single camera to solve the problem of Simultaneous
Localization And Mapping in dynamic environments
obtaining, at the same time, the estimation of the moving objects
trajectories. Specifically, we show that it is possible to segment
the features belonging to independently moving objects from a
moving camera using a MonoSLAM algorithm together with
a Bearing-Only Tracker. The idea is to exchange between two
parallel working systems, i.e. the SLAM filter and the bearingonly
tracker, information about the pose of the camera and
the motion of the feature to improve the robustness of the
SLAM algorithm and maintain a consistent estimation of both
the pose, the map, and the features trajectories. Experiments in
simulated and real environments substantiate that the proposed
technique is able to maintain consistent estimations in a fast
and robust way suitable for a real-time application, even in
situations where classical MonoSLAM algorithms are deemed
to fail.
- Title: Optimal Metric SLAM for Autonomous Navigation Assistance
Authors: P.F. Alcantarilla, I. Parra, L.M. Bergasa
Paper,
Presentation,
video1,
video2
Abstract
In this paper we present a 6DOF metric
SLAM system for outdoor enviroments using a stereo
camera, mounted next to the rear view mirror, as
the only sensor. By means of SLAM the vehicle
global position and a sparse map of natural landmarks
are both estimated at the same time. The system
combines both bearing and depth information using
two di
erent types of feature parametrization: inverse
depth and 3D. Through this approach near and far
features can be mapped, providing orientation and
depth information respectively. Natural landmarks
are extracted from the image and are stored as 3D or
inverse depth points, depending on a depth thresh-
old. At the moment each landmark is initialized, the
normal of the patch surface is computed using the
information of the stereo pair. In order to improve
long-term tracking a 2D warping is done considering
the normal vector information of each patch. This
Visual SLAM system is focused on the localization of
a vehicle in outdoor urban environments and can be
fused with other cheap sensors such as GPS, so as to
produce accurate estimations of vehicle's localization
in a road. Some experimental results under outdoor
environments and conclusions are presented.
Session IV: SLAM, Localization, Reconstruction 15:50-17:50
Chairman: Martin Adams and Sukhan Lee
- Title: Detection Likelihoods for Safer Occupancy Mapping (invited paper)
Authors: John Mullane, Martin Adams, Wijerupage Sardha Wijesoma
Paper,
Presentation,
video1,
video2
Abstract
Typical autonomous navigation algorithms model mobile robot
exteroceptive sensor readings as being corrupted by noise in range and
bearing space only. This implies that spurious sensor (typically range)
readings, which commonly result in dynamic environments, are modeled
with probability density functions within the Cartesian space of the map
to be estimated. This paper shows that many sensors and feature
detection algorithms often produce false alarms and/or missed detections
in environments of high clutter, placing the very existence of estimated
features into question. Hence, the measurement space is redefined in
this paper so that theoretically accurate and state dependent
measurement likelihoods can be obtained and used in the estimation of
feature existence and location certainty. This presentation applies this
detection likelihood framework in complex outdoor environments, using
millimetre wave radar, for autonomous navigation and mapping. Results
are demonstrated which show the higher success rate of the proposed
algorithm, in comparison with standard occupancy mapping algorithms, in
situations of high clutter and missed detections.
- Title: Experimental Comparison of Bayesian Outdoor Vehicle Localization Filters
Authors: Alexandre N. Ndjeng, Dominique Gruyer, Alain Lambert, Sébastien Glaser, Benjamin Mourllion
Paper,
Presentation
Abstract
Localizing a vehicle consists in estimating
its state by merging data from proprioceptive sensors
(inertial measurement unit, gyrometer, odometer, etc.)
and exteroceptive sensors (GPS sensor). A well known
solution in state estimation is provided by the Kalman
filter. But, due to the presence of nonlinearities, the
Kalman estimator is applicable only through some
alternatives among which the Extended Kalman filter
(EKF), the Unscented Kalman Filter (UKF) and the
Divided Differences of 1st and 2nd order (DD1 and
DD2). We have compared these filters using the same
experimental data. The results obtained aim to rank
these approaches by their performances in terms of
accuracy, confidence and consistency.
- Title: Predictive Lane Detection for Simultaneous Road Geometry Estimation and Vehicle Localization
Authors: Chenhao Wang, Zhencheng Hu, Tomoki Maeda, Naoko Hamada, and Keiichi Uchimura
Paper,
Presentation,
video1
Abstract
This paper describes a predictive lane detection method with assistance of road geometry data from digital road map to simultaneously estimate road shape
and vehicle localization. In our approach, visual information is not the only source to detect lane and estimate road parameters, the road geometry information
derived from digital road map has also been providing important predictive cues for lane detection. Comparing with the conventional vision-only based approaches,
our system is able to provide more reliable and stable road geometry estimation result. In addition, a precise longitudinal localization can also be achieved through
the piecewise polynomial matching algorithm. Simulative and real road tests under various environmental conditions have shown the effectiveness of the proposed method.
- Title: Cognitive Localization of 3D Objects Symbolically Given Navigational Cues (invited paper)
Authors: Sukhan Lee, Hyunjun Kim, Zhaojin Lu, and Harry Hung
Paper,
Presentation,
video1,
video2
Abstract
The cyber transportation as a means of autonomous individual public taxi service requires, for its robotic cabs
referred to here as cyber cabs, human-like capabilities of understanding traffic signals and negotiating with
pedestrians and other traffics, and of taking care of various local variations, anomalies, and hazards, as
well as of being ready for service at any locations conveniently set for the users regardless whether they are
inside or outside a building. As such, a cyber cab is desired to have human-like visual capabilities of
understanding 3D scenes, not only with the capability of recognizing humans, objects and artifacts, but
also with the capability of modeling 3D environment or workspace in terms of its configuration and context,
that can be well integrated with the conventional navigational sensing modalities. This paper reports a
progress in this direction of research for cyber trasportation, including 1) the robust recognition
of 3D objects by integrating the conventional engineering approach to vision processes with such cognitive
processes as evidence selection and collection, focus of attention, probabilistic multi-evidence fusion
and incorporation of visual context, 2) the real-time 3D workspace modeling with the identification of
global geometric configurations and the approximate representation of 3D objects based on voxels and
volume primitives, and 3) the modeling by categrization based on the ontology based generic knowledge in DB.
Some experimental results are shown.
- Title: Laser scaner based SLAM in real road and traffic environment
Authors: Olivier Garcia-Favrot, Michel Parent
Paper,
Presentation
Abstract
- In this paper we will present a SLAM
algorithm we have recently developed for our
needs in autonomous automotive applications.
Our approach has the particularity of making use
exclusively of laser scanners to achieve our goals
without using any other type of sensors or source
of information. We concentrated on developing a
self-contained system that could be placed on any
kind of mobile platform and work in any kind of
dynamic environment; this is why too at this
point our approach does not make use of any
model of the vehicle. Our SLAM system has been
tested with success both on a car at full speed on
a road and a human evolving indoors. We will
present here the challenges we face that pushed
us to develop the algorithm, the solutions we are
exploring, discuss experimental results and
suggest areas of future work.
|
