Welcome to Introduction to Self-Driving Cars, the first course in University of Toronto’s Self-Driving Cars Specialization. This course will introduce you to the terminology, design considerations and safety assessment of self-driving cars. By the end of this course, you will be able to: - Understand commonly used hardware used for self-driving cars - Identify the main components of the self-driving software stack - Program vehicle modelling and control - Analyze the safety frameworks and current industry practices for vehicle development For the final project in this course, you will develop control code to navigate a self-driving car around a racetrack in the CARLA simulation environment. You will construct longitudinal and lateral dynamic models for a vehicle and create controllers that regulate speed and path tracking performance using Python. You’ll test the limits of your control design and learn the challenges inherent in driving at the limit of vehicle performance. This is an advanced course, intended for learners with a background in mechanical engineering, computer and electrical engineering, or robotics. To succeed in this course, you should have programming experience in Python 3.0, familiarity with Linear Algebra (matrices, vectors, matrix multiplication, rank, Eigenvalues and vectors and inverses), Statistics (Gaussian probability distributions), Calculus and Physics (forces, moments, inertia, Newton's Laws). You will also need certain hardware and software specifications in order to effectively run the CARLA simulator: Windows 7 64-bit (or later) or Ubuntu 16.04 (or later), Quad-core Intel or AMD processor (2.5 GHz or faster), NVIDIA GeForce 470 GTX or AMD Radeon 6870 HD series card or higher, 8 GB RAM, and OpenGL 3 or greater (for Linux computers).
Lesson 2: Hardware Configuration Design
Loading...
Do curso por University of Toronto
Introduction to Self-Driving Cars
129 classificações
Na lição
Module 2: Self-Driving Hardware and Software Architectures
System architectures for self-driving vehicles are extremely diverse, as no standardized solution has yet emerged. This module describes both the hardware and software architectures commonly used and some of the tradeoffs in terms of cost, reliability, performance and complexity that constrain autonomous vehicle design.
Conheça os instrutores
Steven WaslanderAssociate Professor
Aerospace Studies
Jonathan KellyAssistant Professor
Aerospace Studies
Welcome to the second video of this week.
We started out this week by covering the various kinds of
sensors most commonly used for perception.
Now, let's learn how to place
these sensors to aggregate a complete view of the environment.
In this video, we will cover
the configuration design to meet sensor coverage needs for an autonomous driving car.
We will do this by going through two common scenarios,
driving on a highway and driving in an urban environment.
After analyzing these scenarios,
we will lay out the overall coverage requirements
and discuss some issues with the design.
Let's begin by recalling the most commonly available sensors from our last video.
These are the camera for appearance input.
The stereo camera for depth information,
lidar for all whether 3D input,
radar for object detection,
ultrasonic for short-range 3D input and
GNSS/IMU data and wheel odometry for ego state estimation.
Also, remember that all of these sensors come in
different configurations and have
different ranges in fields of view over which they can sense.
They have some resolution that depends on the instrument specifics and field of view.
Before we move to discussing coverage,
let's define the deceleration rates we're willing to accept for
driving which will drive the detection ranges needed for our sensors.
Aggressive deceleration are set at five meters per second squared which is roughly
the deceleration you experience when you slam the brakes hard
and try to stop abruptly in case of an emergency.
Normal decelerations are set to two meters per second squared,
which is reasonably comfortable while still allowing the car to come to a stop quickly.
Given a constant deceleration our braking distance d can be computed as follows.
D is equal to v squared over 2a,
where V is the vehicle velocity and a is its rate of deceleration.
We can also factor in reaction time of the system and road surface friction limits,
but we'll keep things simple in this discussion.
Let's talk about coverage now.
The question we want to answer is where should we place
our sensors so that we have sufficient input for our driving task?
Practically speaking, we want our sensors to capture the ODD we
have in mind or the ODD our system can produce decisions for.
We should be able to provide all of the decisions with sufficient input.
There can be so many possible scenarios in driving but we'll look at
just two common scenarios to see how the requirements drive our sensor selection.
Will look at highway and urban driving.
Let's think about these two situations briefly.
For a divided highway,
we have fast moving traffic, usually high volume,
and quite a few lanes to monitor,
but all vehicles are moving in the same direction.
The other highlight of driving on a highway setting is that there are
fewer and gradual curves and we have exits and merges to consider as well.
On the other hand, in the urban situation we'll consider,
we have moderate volume and moderate speed traffic with
fewer lanes but with traffic moving in all directions especially through intersections.
Let's start with the highway setting.
We can break down the highway setting into three basic maneuver needs.
We may need to hit the brakes hard if there's an emergency situation.
We need to maintain a steady speed matching the flow of traffic
around us and we might need to change lanes.
In the case of an emergency stop,
if there is a blockage on our road we want to stop in time.
So, applying our stopping distance equation longitudinally,
we need to be able to sense about a 110 meters in front of us assuming
a highway speed of a 120 kilometers and aggressive deceleration.
Most self-driving systems aim for sensing ranges of
a 150 to 200 meters in front of the vehicle as a result.
Similarly, to avoid lateral collision or to
change lanes to avoid hitting an obstacle in our lane,
we need to be able to sense at least our adjacent lanes,
which are 3.7 meters wide in North America.
To maintain speed during vehicle following,
we need to sense the vehicle in our own lane.
Both their relative position and the speed are
important to maintain a safe following distance.
This is usually defined in units of time for
human drivers and set to two seconds in nominal conditions.
It can also be assessed using aggressive deceleration of
the lead vehicle and the reaction time from our ego vehicle.
So, at a 120 kilometers per hour,
relative position and speed measurements to a range of
165 meters are needed and typical systems use 100 meters for this requirement.
Laterally, we need to know what's happening anywhere in our adjacent lanes in case
another vehicles seeks to merge into our lane or we need to merge with other traffic.
A wide 160 to 180 degree field of view is required to track
adjacent lanes and a range of 40 to 60 meters is needed to find space between vehicles.
Finally, let's discuss the lane change maneuver and consider the following scenario.
Suppose we want to move to the adjacent lane,
longitudinally we need to look forward,
so we are a safe distance from the leading vehicle and we also need to look behind just
to see what the rear vehicles are doing and laterally it's a bit more complicated.
We may need to look beyond just the adjacent lanes.
For example, what if a vehicle attempts
to maneuver into the adjacent lane at the same time as we do?
We'll need to coordinate our lane change room maneuvers so we don't crash.
The sensor requirements for lane changes are roughly
equivalent to those in the maintain speed scenario.
As both need to manage vehicles in front of and
behind the ego vehicle as well as to each side.
Overall, this gives us the picture for
coverage requirements for the highway driving scenario.
We need longitudinal sensors and lateral sensors and
both wide field of view and narrow field of view sensors to do these three maneuvers,
the emergency stop, maintaining speed and changing lanes.
Already from this small set of ODD requirements we
see a large variety of sensor requirements that arise.
Let's discuss the urban scenario next.
The urban scenario as we discussed before is a moderate volume,
moderate traffic scenario with fewer lanes on
the highway case but with the added complexity of pedestrians.
There are six types of basic maneuvers here.
Obviously, we can still perform emergency stop,
maintain speed and lane changes but we
also have scenarios such as overtaking a parked car,
left and right turns at intersections and
more complex maneuvers through intersections such as roundabouts.
In fact, for the first three basic maneuvers,
the coverage analysis is pretty much the same as
the highway analysis but since we are not moving as quickly,
we don't need the same extent for our long-range sensing.
Let's discuss the overtake maneuver next.
More specifically, consider a case where you have to overtake a parked car.
Longitudinally, we definitely need to sense
the parked car as well as look for oncoming traffic.
So, we need both sensors,
wide short-range sensors to detect the parked car and
narrow long-range sensors to identify if oncoming traffic is approaching.
And laterally, we'll need to observe beyond
the adjacent lanes for merging vehicles as we did in the highway case.
For intersections, we need to have near
omni-directional sensing for all kinds of movements that can occur.
Approaching vehicles, nearby pedestrians,
doing turns and much more.
Finally, for roundabouts we need
a wide-range, short distance sensor laterally since the traffic is slow
but we also need a wide-range short distance sensor
longitudinally because of how movement around the roundabout occurs.
We need to sense all of the incoming traffic
flowing through the roundabout to make proper decisions.
And so we end up with this overall coverage diagram for the urban case.
The main difference with respect to
highway coverage is because of the sensing we require for
movement at intersections and at roundabouts and for the overtaking maneuver.
In fact, the highway case is almost entirely covered by the urban requirements.
Let's summarize the coverage analysis.
For all of the maneuvers we do,
we need long range sensors which typically have shorter angular field of view
and wide angular field of view sensors which
typically have medium to short-range sensing.
As the scenarios become more complex,
we saw the need for full 360 degrees sensor coverage on
the short scale out to
about 50 meters and much longer range requirements in the longitudinal direction.
We can also add even shorter range sensors like sonar which are useful in
parking scenarios and so in
the end our sensor configuration looks something like this diagram.
To summarize, our choice of sensors should be driven by
the requirements of the maneuvers we want to execute and
it should include both long-range sensors for
longitudinal dangers and wide field of view sensors for omnidirectional perception.
The final choice of configurations also
depends on our requirements for operating conditions,
sensor redundancy due to failures and on budget.
There is no single answer to which sensors are needed for a self-driving car.
In this video, you learned how to select a hardware configuration by doing
sensor coverage analysis both for
longitudinal and lateral cases for highway and urban driving.
In the next video,
we'll study a modular software architecture for
a typical autonomous driving stack. See you then.
Explore nosso catálogo
Registre-se gratuitamente e obtenha recomendações, atualizações e ofertas personalizadas.
O Coursera proporciona acesso universal à melhor educação do mundo fazendo parcerias com as melhores universidades e organizações para oferecer cursos on-line.
© 2019 Coursera Inc. Todos os direitos reservados.
