Focus on Dynamic Systems & Control

CONNECTED COOPERATIVE CONTROL

OF AUTONOMOUS VEHICLES
DURING UNEXPECTED

Downloaded from http://asmedigitalcollection.asme.org/memagazineselect/article-pdf/139/12/S3/6360487/me-2017-dec7.pdf by guest on 08 February 2021

A
utonomous driving in urban
environments has been of
great interest to researchers
due in part to the high density

SITUATIONS
of vehicles and various area-
specific traffic rules that must
be obeyed. The DARPA Urban Challenge
[1], and more recently the V-Charge
Project catalyzed the launch of research
efforts into autonomous driving on urban BY HANS ANDERSEN DANIELA RUS
roads for numerous organizations. Re- DEPARTMENT OF COMPUTER SCIENCE AND ARTIFICIAL
MECHANICAL ENGINEERING INTELLIGENCE LABORATORY
ferring to Figure 1, the problem of urban
NATIONAL UNIVERSITY MASSACHUSETTS INSTITUTE
driving is both interesting and difficult OF SINGAPORE OF TECHNOLOGY
because it encompasses both increased
operating speeds of autonomous ve- XIAOTONG SHEN MARCELO H. ANG JR.
hicles as well as increased environmen- YOU HONG ENG DEPARTMENT OF
tal complexity. A mature solution in one SINGAPORE-MIT ALLIANCE FOR MECHANICAL ENGINEERING
environment may not work in another RESEARCH AND TECHNOLOGY NATIONAL UNIVERSITY OF SINGAPORE
due to different traffic rules and human
driving characteristics that are unique in Perception is the ability of an autonomous system to extract relevant
each urban area. A particularly difficult knowledge and understanding about itself and its environment through
internal sensing and external environmental sensing. Internal sensing
problem arises when unexpected situa-
is essentially to observe the states of current sensors, switches, and ac-
tions happen during the autonomous run, tuators, which are mainly used for self-diagnosis. The external sensing
and may require the unmanned system includes estimation of the current location, map features, and dynamic
to break the corresponding traffic rule in objects, which are used for localization, mapping, and obstacle detec-
tion, respectively. The detected obstacles are considered in both path
order to progress along its own course.
planning and speed control.
Vehicle-to-Vehicle (V2V) communication Planning for autonomous driving is usually performed in a hierarchi-
offers the promise of enhancements on cal manner. The mission planner (or route planner) considers high-
both urban driving fronts, especially when level objectives, such as assignment of pickup/drop-ŏ tasks and which
roads should be taken to achieve the tasks. The behavioral planner
faced with unexpected situations.
(or decision-maker) makes ad-hoc decisions to properly interact with
An overview of autonomous vehicle other agents and follow rules restrictions, and thereby generates local
software architecture [2] is shown in objectives, e.g., change lanes, overtake, or proceed through an intersec-
Figure 2. The subsystems of an autono- tion. The motion planner generates a locally-optimal path that avoids
unexpected obstacles. The planned path is then fed into the motion
mous vehicle can be broadly grouped
control module.
into three categories: perception, plan- The motion control module consists of several subsystems. The lon-
ning and control. gitudinal controller outputs brake or throttle signals to the actuation

DECEMBER 2017 3
system so that the speed of the vehicle tracks the desired speed. A few dĭerent approaches have been proposed in recent litera-
The lateral controller outputs a steering signal to the actuation ture to handle this kind of unexpected dilemma. Sampling-based
system so that the vehicle follows the desired path. In case of any methods, such as RRT and its variants, are popular for trajec-
emergency situations, the emergency module will be enabled to tory planning. 2ne notable variant, Minimum Violation RRT
stop the vehicle appropriately. (MVRRT ), has been proposed by Reyes Castro, et. al >@. The
Autonomous driving on urban roads has seen tremendous prog- authors express trȧc rules as formulas using /inear Temporal

FIGURE 1 Complexity
and operating velocity
for various driving
scenarios. Source: [1]

Downloaded from http://asmedigitalcollection.asme.org/memagazineselect/article-pdf/139/12/S3/6360487/me-2017-dec7.pdf by guest on 08 February 2021

FIGURE 2 A typical
autonomous vehicle
system overview,
highlighting core com-
petencies. Source: [1]

ress in recent years, with several commercial entities pushing the /ogic (/T/), and propose an incremental algorithm to generate a
research boundaries alongside academia. *oogle has perhaps the trajectory of a dynamical system that systematically picks which
most experience in the area, having tested its Àeet of autonomous safety rules to violate and minimizes the level of risk involved. The
vehicles for more than 2 million miles >@. Tesla is early to market system assumes a static environment, and that the environment
their work, having already provided an autopilot feature in their is known a priori. The proposed system also relies on a carefully
2 Model S cars. Uber¶s mobility service has grown to upset the designed set of rules and formulations.
taxi markets in numerous cities worldwide, and has furthermore *uo, et al. >@ proposed a solution to circumventing the illegally
recently indicated plans to eventually replace all their human-driv- parked vehicle by ¿nding a lead vehicle in the ego lane and
en Àeet with self-driving cars. Nutonomy is the ¿rst company in the following its behavior to generate a trajectory that is based on a
world to introduce autonomous taxi service, which hit the roads of cubic spline model with mass-spring-damper system. However,
Singapore in August 2 >@. Nutonomy¶s success can also be at- this approach may fail if there are no leading vehicles in the ego
tributed to the Singapore *overnment¶s initiative in opening some lane or if the intention of the vehicle is unknown, as the urban
of the roads in one-north (Figure 3), a technology business district trȧc rules can be complicated and very dynamic.
for autonomous vehicle testing. /ee and Seo >@ have proposed another learning-based method
However, all of the above-mentioned companies have reported for such circumstances. They proposed a framework based on
accidents while driving autonomously. A preliminary analysis in inverse reinforcement learning and a *aussian process. Real-world
2 by Schoettle and Sivak >@ has shown that autonomous ve- data collected from expert drivers are used to train a trajectory
hicles have a higher crash rate per million miles traveled compared generator. Using the pre-trained weight, an optimal trajectory can
to conventional vehicles, and similar patterns were evident for be evaluated online. This approach also relies on manually de¿ned
injuries per million miles traveled and for injuries per crash. The and engineered features that have to be carefully chosen. The
report also concluded that none of the accidents reported thus far method also sŭers from discretization error due to discontinuity
has been the fault of the autonomous vehicles, as their vehicles in the problem formulation and training. In general, learning-
have been programmed to follow the trȧc rules conservatively. based motion planning methods often act as black boxes that are
very di̇cult to systematically analyze and therefore prove safety.
UNEXPECTED SITUATIONS /earning-based methods also rely on availability of valid expert

R eacting to potentially hazardous unexpected situations is

one of the key issues in autonomous driving in urban environ-
ments. An example scenario that we encounter very frequently
data and feature engineering. Acceptable driving styles under
unexpected situations can dĭer from one place to another, and
therefore a network that has been trained under one circumstance
during our autonomous vehicle deployment at the 2ne-north area may not be applicable in the other.
in Singapore is depicted in Figure 4. In this scenario, a car is Such problem is often formulated as a constrained optimization
illegally parked on the vehicle¶s ego lane, and therefore has to be problem and the locally-optimal solution to the problem is computed
overtaken. In this case, a human driver may have to move slightly with a receding horizon. This controller is referred to in the literature
into the opposite lane in order to clearly see in front of a car ahead. as a Model Predictive Controller (MPC). Compared to learning-
2nce he has gathered enough information about the road ahead, based methods, MPC requires more in-depth understanding of the
then he can safely overtake. However, as this is two-way trȧc, the problem, and accurate problem modeling and formulation. How-
overtaking implies that the vehicle invades to the opposite lane, ever, in contrast to learning methods, there is a huge literature on
and therefore will take the trȧc head-on, causing a safety hazard. the analytical aspects of the optimization problem, and therefore it is

4 DECEMBER 2017
 Focus on Dynamic Systems & Control
possible to design a controller that balances safety and complexity. the obstacles. However, there are risks associated with the limited
MPC has a few other attractive features. First, it is possible to perception range of the on-board sensors of the vehicle. These risks
intuitively incorporate vehicle dynamics into the problem formula- can be mitigated by having an inter-vehicle communication system,
tion. Second, the problem can be formulated in continuous time, which will be discussed next.
and therefore does not possess the problems that probabilistic
motion planning methods possess, including inherent inaccuracy CONNECTED VEHICLES
due to discretization limits, and the computational complexity
that rises exponentially as the dimensionality of the planning state
space increases.
C ooperation between multiple autonomous vehicles (AVs) is
possible with the development of vehicular communication. In
particular, state estimation can be improved with multiple sources of
Probabilistic motion planning methods’ main strength is its information gathered from dĭerent vehicles. Cooperative state esti-
probabilistic completeness, and global optimality. However, due to mation can also improve robustness against communication failure.
the limitation in the sensor range and the uncertain nature of driv- With future trajectories shared among nearby vehicles, the motion
ing in an urban environment, re-planning with a receding horizon can be coordinated to make navigation safer and smoother for AVs.

Downloaded from http://asmedigitalcollection.asme.org/memagazineselect/article-pdf/139/12/S3/6360487/me-2017-dec7.pdf by guest on 08 February 2021

is always necessary, and therefore it may be more practical to plan
a locally-optimal solution within the prediction horizon. VEHICULAR COMMUNICATION
Researchers have also approached this problem from the philo-
sophical point of view, which argues whether autonomous vehicles
have to be programmed to take the action that causes the least
V ehicular communication technology has been progressing
rapidly, enabling connection between vehicles via wireless
networks. The bandwidth and range of wireless communication are
damage. Gerdes and Thornton [9] attempt to answer the ethical increasing rapidly while the latency is being signi¿cantly reduced.
question for handling such dilemmas by formulating the motion For example, the communication range of Dedicated Short Range
planning as an MPC problem. They argue that ethical autonomous Communications (DSRC) can be up to one kilometer, allowing a
vehicles must obey trȧc rules, except where obeying the trȧc vehicle to connect to nearby vehicles even beyond line-of-sight
rules could cause a collision with human agents, other vehicles or and ¿eld-of-view. Furthermore, the information can be relayed
the environment. Therefore, trȧc rules have to be formulated as a and multi-hop connections are possible, which can signi¿cantly
cost term in the MPC formulation. increase the connectivity. For vehicular communication, the IEEE
In recent work [10], we have formulated the problem of overtak- 802.11p standard has been designed to allow information exchange
ing an illegally parked vehicle on a 2-way street as an MPC prob- between high-speed cars, and between vehicles and roadside infra-
lem. Referring to Figure 5, unexpected objects on the ego lane will structure. Other wireless communication technologies, such as 3G,
cause occlusion and therefore the vehicle has to move out of its ego 4G and WiFi, are also suggested in [11].

FIGURE 3 Autonomous
vehicle testing area in
One-north, Singapore.
Image: Delphi.

FIGURE 4 Unexpected
static obstacle in the
form of an illegally-
parked car on a two-
way street.
Image: Hans Andersen.

lane to gather su̇cient information before making the decision COOPERATIVE LOCALIZATION
whether to overtake the obstacle or not. We have observed the
following behavior of human drivers facing the described scenario:
they will ¿rst decelerate the vehicle, and move closer to the center
G lobal Positioning System (GPS) is a widely-used method for
estimating a vehicle’s location, however, it is generally unavail-
able or unreliable due to signal obstruction or multi-path ĕects,
of the lane and assess the trȧc on the opposite lane as well as especially in urban environments. Cooperative information sharing
the distance that the driver has to overtake, before ¿nally overtak- and fusion enables signi¿cant improvement in vehicle localization,
ing the obstacle and merging back to the ego lane. Based on this e.g., by installing transmitters in the infrastructure, correction mes-
observation, we have designed a behavior planner with costs and sages can be shared so as to improve the estimation accuracy. In
constraints of the MPC problem. In contrast to previous works, we order to reduce the common GPS bias, the GPS coordinates can be
also consider visibility maximization (blind spot minimization), to shared with neighboring vehicles through vehicle to vehicle (V2V)
generate overtaking trajectories that take into account the percep- communication and recti¿ed by applying a constraint that the group
tion limitations of the ego vehicle. of vehicles must all reside on the road. Usually, a digital map, i.e.,
Simulation results have shown that the proposed method is the road network, is used for group map matching, however other
capable of making a safe decision when deciding and overtaking approaches such as pairwise map merging using Simultaneous Lo-

DECEMBER 2017 5
 calization and Mapping (SLAM) methods [12] can also be utilized for
REFERENCES estimating the relative pose between two vehicles. Relative observa-
1 S. Pendleton, H. Andersen, X. Du, X. Shen, M. Meghjani, Y. H. Eng, D. Rus, and M. tions are commonly used for cooperative localization, which can be
H. Ang, “Perception, Planning, Control, and Coordination for Autonomous Vehicles,” categorized into four groups: relative range, relative bearing, relative
Machines, vol. 5, no. 1, 2017, p. 6. position, and relative pose. Some sensors can only give the range
2 S. Buehler, Martin; Iagnemma, Karl; Singh, Ed., The Darpa Urban Challenge information while others can only give bearing information. For
Autonomous Vehicle in City Traffic. Berlin: Springer, 2009. instance, acoustic sensors can only measure the relative distance by
3 T. Higgins, “Google’s Self-Driving Car Program Odometer Reaches 2 Million Miles measuring the Time of Arrival (TOA), while the monocular camera
- WSJ.” [Online]. Available: https://www.wsj.com/articles/googles-self-driving-car- can only measure the bearing angle. If the sensor can measure both
program-odometer-reaches-2-million-miles-1475683321. [Accessed: 04-Oct-2017]. TOA and Angle of Arrival (AOA), it can provide the relative position.
4 J. M. Watts, “World’s First Self-Driving Taxis Hit the Road in Singapore - WSJ.” After the shared information has been received, the remaining
[Online]. Available: https://www.wsj.com/articles/worlds-first-self-driving-taxis-
two steps of cooperative localization are data association and data
hit-the-road-in-singapore-1472102747. [Accessed: 04-Oct-2017].
fusion. Most of cooperative localization is performed in simula-

Downloaded from http://asmedigitalcollection.asme.org/memagazineselect/article-pdf/139/12/S3/6360487/me-2017-dec7.pdf by guest on 08 February 2021

5 B. Schoettle and M. Sivak, “A Preliminary Analysis of Real-World Crashes Involving
tion, where vehicle identities are assumed to be known. The vehicle
self-Driving Vehicles,” Ann Arbor, MIchigan, USA, 2015.
identity problem is solved by using a distinct transmitting signal
6 L. I. R. Castro, P. Chaudhari, J. Tumova, S. Karaman, E. Frazzoli, and D. Rus,
“Incremental sampling-based algorithm for minimum-violation motion planning,” [13]. The data association is a challenging problem due to com-
in 52nd IEEE Conference on Decision and Control, 2013, pp. 3217–3224. binatorial explosion. Some methods use topology information or
7 C. Guo, K. Kidono, and M. Ogawa, “Learning-based trajectory generation for symmetric measurement equation (SME) to avoid data association.
intelligent vehicles in urban environment,” IEEE Intell. Veh. Symp. Proc., vol. 2016– Other methods use PHD ¿lter, nearest neighbors assignment with
Augus, no. Iv, 2016, pp. 1236–1241. validation gate, and multiple hypotheses registration. For the data
8 S. H. Lee and S. W. Seo, “A learning-based framework for handling dilemmas in fusion, various methods are proposed, such as standard Kalman
urban automated driving,” in 2017 IEEE International Conference on Robotics and ¿lter, cubature Kalman ¿lter, Covariance Intersection ¿lter, particle
Automation (ICRA), 2017, pp. 1436–1442. ¿lter, factor graph optimization, maximum likelihood estimation
9 J. C. Gerdes and S. M. Thornton, “Implementable Ethics for Autonomous Vehicles,” (MLE), and Maximum A-Posteriori Estimation (MAP).
in Autonomous Driving: Technical, Legal and Social Aspects, M. Maurer, J. C. Gerdes, B. The open problems of multi-vehicle cooperative localization
Lenz, and H. Winner, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2016, pp. should at least include the following: communication delay and
87–102.
failure [13]; data bandwidth and cluttered environment [14]; robust
10 H. Andersen, W. Schwarting, F. Naser, Y. H. Eng, D. Rus, and J. Alonso-mora, data association and scalability [15].
“Trajectory Optimization for Autonomous Overtaking with Visibility Maximization,”
in IEEE International Conference on Intelligent Transportation Systems (ITSC), 2017.
COOPERATIVE CONTROL

P
11 X. Shen, Z. J. Chong, S. Pendleton, G. M. J. Fu, B. Qin, E. Frazzoli, and M. H. Ang,
“Teleoperation of on-road vehicles via immersive telepresence using off-the-shelf lanned future trajectories can also be shared so that the pre-
components,” in Intelligent Autonomous Systems 13, Springer, 2016, pp. 1419–1433. diction of cooperating vehicles’ future positions can be better
12 H. Li, M. Tsukada, F. Nashashibi, and M. Parent, “Multivehicle cooperative local facilitated. Potential motion conÀicts can then be identi¿ed and
mapping: A methodology based on occupancy grid map merging,” IEEE Trans. Intell. mitigated with motion coordination algorithms, which can guaran-
Transp. Syst., vol. 15, no. 5, 2014, pp. 2089–2100. tee that decisions are jointly feasible.
13 J. M. Walls, A. G. Cunningham, and R. M. Eustice, “Cooperative localization by With future trajectories shared among vehicles via vehicle to
factor composition over a faulty low-bandwidth communication channel,” Proc. - vehicle (V2V) communication, the collective vehicle motion can be
IEEE Int. Conf. Robot. Autom., vol. 2015–June, no. June, 2015, pp. 401–408. coordinated in an optimal way to avoid conÀicts. Multi-robot mo-
14 D. Gulati, F. Zhang, D. Clarke, and A. Knoll, “Graph-Based Cooperative tion planning has been studied extensively to take into account the
localization using symmetric measurement equations,” Sensors (Switzerland), vol. paths of other robots so as to avoid any possible collision, congestion
17, no. 6, 2017, pp. 1–16.
or deadlock. A wide variety of methods have been proposed in the
15 X. Shen, S. Pendleton, and M. H. Ang Jr, “Scalable cooperative localization with literature, which is often categorized along the spectrum between
minimal sensor configuration,” in Distributed Autonomous Robotic Systems, Springer,
2016, pp. 89–104. centralized and decoupled planning. Centralized approaches plan
the path in the composite con¿guration space that is formed by the
16 M. Cap, P. Novak, A. Kleiner, and M. Selecky, “Prioritized Planning Algorithms for
Trajectory Coordination of Multiple Mobile Robots,” IEEE Trans. Autom. Sci. Eng., vol. Cartesian product of con¿guration spaces of individual robots and
12, no. 3, 2015, pp. 835–849. then extracts the trajectories for the individuals to execute. Proba-
17 M. Saha and P. Isto, “Multi-robot motion planning by incremental coordination,” bilistic motion planning algorithms, such as A*, D* and RRT*, can
IEEE Int. Conf. Intell. Robot. Syst., 2006, pp. 5960–5963. be leveraged to ensure completeness and optimality. The decoupled
18 X. Shen, Z. J. Chong, S. Pendleton, W. Liu, B. Qin, G. M. J. Fu, and M. H. Ang, planning can be further classi¿ed into two, namely prioritized plan-
“Multi-vehicle motion coordination using V2V communication,” in 2015 IEEE ning and path-velocity planning. The prioritized planning method
Intelligent Vehicles Symposium (IV), 2015, pp. 1334–1341. plans the path sequentially, according to the prede¿ned or online
computed priorities, and robots with planned paths are regarded as
dynamic obstacles in the con¿guration-time space for the remain-
ACKNOWLEDGEMENTS ing robots to avoid. Much of the related research work has focused
on the assignment of priorities to improve the quality of the solution
This research was supported by the Future Urban Mobility [16]. The path-velocity planning method plans the path concurrently
project of the Singapore-MIT Alliance for Research and
while ignoring the mutual collisions in the ¿rst phase and resolves
Technology (SMART) Centre, with funding from Singapore’s
the conÀicts by velocity planning in the second phase. A hybrid of
National Research Foundation (NRF).
prioritized planning and path-velocity planning is introduced in [17]

6 DECEMBER 2017
 Focus on Dynamic Systems & Control
where motion coordination is conducted in an incremental manner.
Nonetheless, the decoupled planning sacri¿ces the completeness and
optimality for ėciency and applicability.
While there are many multi-robot/multi-vehicle motion plan-
ning algorithms available, only some are actually applicable in
multi-vehicle motion coordination. There is some uniqueness to FIGURE 5
the multi-vehicle motion coordination problem. The following four Complete
occlusion (top)
traits are speci¿c to multi-vehicle motion coordination:
and blind
1) The goals are usually not interchangeable for vehicles because spot (bottom)
each vehicle has its own destination, and thus there is no need to of a vehicle
maintain communication connectivity. approaching
2) The vehicles need to stay in the middle of the lane, and thus an obstacle.
Source: [10]
the path is ¿xed in most circumstances.

Downloaded from http://asmedigitalcollection.asme.org/memagazineselect/article-pdf/139/12/S3/6360487/me-2017-dec7.pdf by guest on 08 February 2021

3) The vehicles are usually moving fast and thus communication
latency is a critical variable in collision avoidance.
4) Reverse motion is typically not allowed on the road because of poses a completely dĭerent challenge due to the complexity of the
trȧc rules. trȧc rules as well as unexpected scenarios involved.
An example motion coordination algorithm that considers these Reacting to these scenarios is still a very challenging topic, es-
aspects is proposed in [18]. In the proposed method, V2V com- pecially when the autonomous vehicle has to break trȧc rules, or
munication is combined with graph search in the coordination pick the best of two evils. The ultimate goal of deploying autono-
diagram to resolve conÀicts in future trajectories and minimize the mous vehicles is to provide safe and comfortable mobility, and thus
total waiting time, and plan time-optimal trajectories. it is important to reduce the instances in which the system has to
make such decisions by managing the unexpected risks associated
CONCLUSION with unenforced trȧc rules.

A utonomous vehicles have come a long way from research labs

to nearing full commercialization. However, we believe that
its best days are still ahead. Many modern cars have been adver-
Future research has to address these issues not only by planning
safe behavior and motion, but also harnessing the superhuman
perception that connected vehicles enable. Finally, it is then critical
tised to have autonomous driving capabilities, but these features to carefully integrate all of the software components in the system,
are mostly demonstrated for automated highway driving, and still ensuring that the interactions between dĭerent software compo-
require human attention. Driving autonomously in urban areas nents are meaningful and valid. Q

ABOUT THE AUTHORS

Hans Andersen received gree in industrial and systems the Department of Mechanical Engineering of
B.Eng (Hons) in Mechanical En- engineering from the National the National University of Singapore, where
gineering and M.Sc. in Mecha- University of Singapore in he is currently an Associate Professor, with a
tronics degrees from National 2015. From 2008 to 2015, he Joint Appointment at the Division of Engineer-
University of Singapore in 2013 was a Research Associate with ing and Technology Management. He is also
and 2014, respectively. He is the Acoustic Research Labora- the Acting Director of the Advanced Robotics
currently working toward the Ph.D. degree in tory (ARL), Tropical Marine Science Institute, Centre. His research interests span the areas
the department of mechanical engineering, where he worked on the identification, navi- of robotics, mechatronics, and applications of
National University of Singapore. His research gation and control of unmanned underwater intelligent systems methodologies.
interests include autonomous driving, motion vehicles. He is also interested in the develop-
planning, and model predictive control. ment of new types of actuators for underwater Daniela Rus is the Andrew and
applications. He then joined the Singapore- Erna Viterbi Professor of Electri-
Xiaotong Shen received the MIT Alliance for Research and Technology as cal Engineering and Computer
B.Eng. degree in mechanical a Postdoctoral Associate to work on the self- Science and Director of the
engineering from Harbin In- driving vehicle. Computer Science and Artificial
stitute of Technology, Harbin, Intelligence Laboratory (CSAIL)
China, and the Ph.D. degree in Marcelo H. Ang, Jr. received at MIT. She serves as the Director of the Toyota-
mechanical engineering from his Ph.D. degree in Electrical CSAIL Joint Research Center and is a member
National University of Singapore in 2016. He is Engineering from the University of the science advisory board of the Toyota
currently a Postdoctoral Scholar with the Singa- of Rochester, Rochester, New Research Institute. Rus’ research interests are in
pore-MIT Alliance for Research and Technology. York, in 1988. His work experi- robotics, mobile computing, and data science.
His research interests include cooperative local- ence includes heading the Rus is a Class of 2002 MacArthur Fellow, a fel-
ization, cooperative mapping, and cooperative Technical Training Division of Intel’s Assembly low of ACM, AAAI and IEEE, and a member of
perception for autonomous vehicles. and Test Facility in the Philippines, research the National Academy of Engineering and the
positions at the East West Center in Hawaii and American Academy of Arts and Sciences. She is
You Hong Eng received the B.Eng. (Hons.) and at the Massachusetts Institute of Technology, the recipient of the 2017 Engelberger Robotics
M.Eng. degrees in mechanical engineering and a faculty position as an Assistant Professor Award from the Robotics Industries Associa-
from the Nanyang Technological University, of Electrical Engineering at the University of tion. She earned her PhD in Computer Science
Singapore, in 2006 and 2008, and the Ph.D. de- Rochester, New York. In 1989, Dr. Ang joined from Cornell University.

DECEMBER 2017 7