Research on Perception Gaps in Self-Driving Cars (Autonomous Vehicles)

This report examines the perception challenges of autonomous vehicles, focusing on current limitations and potential research areas for PhD studies. The analysis covers various aspects, including the reliance on sensors, software vulnerabilities, and the impact of environmental conditions on performance. The report highlights the importance of machine perception, fusion algorithms, and the potential for sensor failures, especially in adverse weather. It also discusses the need for continuous software updates and the limitations of current simulation tools. The research identifies gaps in areas such as 3D rendering, sensor noise, and simulator maturity, and proposes several research topics for further investigation, including the future of autonomous vehicles, different aspects of machine perceptions, and the social dilemma of autonomous vehicles.

Running head: AUTONOMOUS VEHICLES
AUTONOMOUS VEHICLES
Name of the Student
Name of the university
Author note

1AUTONOMOUS VEHICLES
Customers throughout the world have a lot of enthusiasm related to the development
of autonomous cars. Autonomous cars do not require any human for their operation and in
this case human intervention can be said to be least or minimal. The advent of autonomous
cars has been taken as a mark that technology has advanced at a rapid rate but the main aspect
in this case is that there are many limitations to the same. The limitations of self-driving cars
will be analyzed in this paper. Autonomous vehicles are proposed to bring in reductions in
the fatalities that take place in roads (Haboucha, Ishaq & Shiftan, 2017). These facilitate the
same by switching control of the critical tasks from the humans to the machines. In order to
realize the safety benefits the concept of machines automatically determining their ideal
behaviour to bring in maximization in performance needs to be understood. This paper will
be discussing on the same.
There have been many researches on the advantages that autonomous vehicles
provide but there are fewer cases where the limitations that it is laced with is mentioned or
given importance to. The development of a new technology can be assessed by perceived
efficacy, ease of use, cost of the same, general attitude of the people towards it and reliability
on the same. Autonomous vehicles do have some unique potential factors that need to be
considered, as enhance functioning and many more (Janai, Güney, Behl & Geiger, 2017). The
technology can be adopted widely depending on the perceptions of the public about the same.
The adoption of the technology by the society can be the only way to judge its success in the
field. Reinforcement learning refers to machine learning that permits the machines as well as
the software agents to determine the profit maximization behaviour on their own. Several real
time systems work at a time to operate an autonomous vehicle. These include route planning,
movement control, localization and environment mapping (Favarò, Eurich & Nader, 2018).
This can be said to be as machine perception that can be defined as ability of computers to
interpret data in a similar manner to that of humans. Software and programming has a vital

2AUTONOMOUS VEHICLES
part to play in this case. The sensory input is made use of by the computer systems for this
purpose.
As per the research done by Litman, (2017), these are expensive and the parts used
are unique and fine in nature. Unlike the present day cars these are not easy to develop as
they require advanced software, high level vehicle parts along with sensors. The goal of
machine perception is to provide these with the ability like the humans to observe, feel as
well as perceive the environment around them and take decisions like their human
counterparts. This can be said to be one of major aspect that can lead towards a future where
machines will replace humans as they will be able to feel and perceive as well. It can be said
that development of software can be attributed for the same.
As per the research done by Cui, Seibold, Stern & Work, (2017), the advantages of
autonomous cars are talked off by everyone but the safety and security issues related to the
same are mostly avoided or neglected. It can be said that though being programmed well
there are high chances that this technology will face some or other glitch that can have drastic
effects and is associated with life risks as well. There can be different reinforcement learning
applications in case of autonomous vehicles which can be related to the riving systems of the
same. The driving behaviour of these autonomous cars in highways can be understood by
using convolution neural networks with Q-learning. Thus autonomous cars are considered as
agents those who are learning to drive safely (Fridman, 2018). The environment to test the
same is created that includes certain fixed agents as well as human drivers. Continuous
updates to the software are a vital thing and the equipments may have faulty codes resulting
from the incomplete updates can be the cause of corrupted programming that can disgrace
from the car from the task that it has been programmed for.

3AUTONOMOUS VEHICLES
As per the research done by Combs, Sandt, Clamann & McDonald, (2019), in the
perception system of the autonomous vehicles cameras are classified as visible or infrared.
The element that is made use of by camera for capturing a scene is called the imaging sensor
and there are two technologies namely Charge-coupled device (CCD) and complementary
metal oxide semiconductor (CMOS) that are made use of. The manufacturing process of
CCD image sensor is expensive and thus they have unique properties including high
quantification energy as well as low noise. CMOS reduces the cost related to manufacturing
but it compromises with the performance. This can have a negative impact on the working of
the system.
As per the research done by Meyer, Becker, Bösch & Axhausen, (2017), fusion
algorithms are made use of for improvising the measurement of multiple sources of data that
are obtained from sensors. Sensory fusion is made use of to measure redundant data reducing
the uncertainty related to measurement of data. This brings in accuracy improvising the
integrity of the system. This helps to deal with faults and helps to improve the fault tolerance
mechanism in the system. Obtaining a classification of these fusion techniques is really
difficult that can be attributed to the multidisciplinary along with the large number of
different case studies as provided in the literature (Lin, Gong, Li & Peeta, 2018).
As per the research done by Hancock, Nourbakhsh & Stewart, (2019), the
autonomous vehicles rely on sensors and failure of the same can pose threats and involves
life risk as well. During sudden changes in the weather conditions there have been instances
of sensor failures and autonomous cars have failed in their operations leading to mishaps.
Here the positioning system can be discussed which is a vital aspect for navigation in these
kind of vehicles. There are many applications using fusion methods with sensors that belong
to positioning as well as environment perception systems (Pendleton et al., 2017). Simulation
along with modelling is considered as essential tools for the purpose of analysis, acquisition,

4AUTONOMOUS VEHICLES
training as well as design in case of the automotive industry which includes the autonomous
vehicles. There has been an increase in the tools for simulation of the autonomous vehicles
and thus it is tough to choose from among all these. Robotics simulators are also been used
for simulating the autonomous vehicles. These robotics simulators need to provide modelling
of all the sensors as well as the actuators that are present in the autonomous vehicles.
Thus from the above discussion it can be said that though autonomous vehicles are a
revolutionary development but the software perceptions related to the same have a vital role
to play. There are high chances that harsh weather conditions affect the sensors which in turn
can be a hindrance to the operation of the vehicle. It can be a challenge for the autonomous
vehicle to face sudden changes in weather or handle hazardous terrains. In order to
successfully implement the autonomous vehicles many things need to be considered such as
3D rendering, sensor noise, license of the simulator, level of maturity of the simulator and
many more such things. The autonomous vehicles are highly dependent on sensors or rather
the feedback obtained from the same thus it can be said that virtualization of sensors is an
essential part to any of the simulation platforms in case of autonomous vehicles. Hence, in
case of autonomous vehicles the software application perception is important and needs to be
discussed on. The technology has its base on the software design and programming thus these
needs to be evaluated.
Research topics
1. Future of autonomous vehicles.
2. Different aspects of machine perceptions.
3. The social dilemma of autonomous vehicles.

5AUTONOMOUS VEHICLES
References
Combs, T. S., Sandt, L. S., Clamann, M. P., & McDonald, N. C. (2019). Automated vehicles
and pedestrian safety: exploring the promise and limits of pedestrian
detection. American journal of preventive medicine, 56(1), 1-7.
Cui, S., Seibold, B., Stern, R., & Work, D. B. (2017, June). Stabilizing traffic flow via a
single autonomous vehicle: Possibilities and limitations. In 2017 IEEE Intelligent
Vehicles Symposium (IV) (pp. 1336-1341). IEEE.
Favarò, F., Eurich, S., & Nader, N. (2018). Autonomous vehicles’ disengagements: Trends,
triggers, and regulatory limitations. Accident Analysis & Prevention, 110, 136-148.
Fridman, L. (2018). Human-centered autonomous vehicle systems: Principles of effective
shared autonomy. arXiv preprint arXiv:1810.01835.
Haboucha, C. J., Ishaq, R., & Shiftan, Y. (2017). User preferences regarding autonomous
vehicles. Transportation Research Part C: Emerging Technologies, 78, 37-49.
Hancock, P. A., Nourbakhsh, I., & Stewart, J. (2019). On the future of transportation in an era
of automated and autonomous vehicles. Proceedings of the National Academy of
Sciences, 116(16), 7684-7691.
Janai, J., Güney, F., Behl, A., & Geiger, A. (2017). Computer vision for autonomous
vehicles: Problems, datasets and state-of-the-art. arXiv preprint arXiv:1704.05519.
Lin, L., Gong, S., Li, T., & Peeta, S. (2018, July). Deep learning-based human-driven vehicle
trajectory prediction and its application for platoon control of connected and
autonomous vehicles. In The Autonomous Vehicles Symposium (Vol. 2018).
Litman, T. (2017). Autonomous vehicle implementation predictions (p. 28). Victoria, Canada:
Victoria Transport Policy Institute.
Meyer, J., Becker, H., Bösch, P. M., & Axhausen, K. W. (2017). Autonomous vehicles: The
next jump in accessibilities?. Research in transportation economics, 62, 80-91.

6AUTONOMOUS VEHICLES
Pendleton, S. D., Andersen, H., Du, X., Shen, X., Meghjani, M., Eng, Y. H., ... & Ang, M. H.
(2017). Perception, planning, control, and coordination for autonomous
vehicles. Machines, 5(1), 6.