Program

The Challenges of Productizing Autonomous Driving

Achieving Mass Commercialization: What will it take for autonomous driving to go mainstream?

Fully autonomous vehicles are coming in the not-so-distant future, and it won’t be long before self-driving cars are available to everyone. But what will it take for autonomous driving to go mainstream? What technology is necessary to make it happen? How can the industry make prices reasonable for the masses? What obstacles stand in our way, and how can we overcome them? In this session, Innoviz CEO, Omer Keilaf, will answer these questions as others as he presents a roadmap for achieving mass commercialization of autonomous vehicles.
 
 

Staying ahead of the curve: Radar and the future of autonomous driving

The autonomous driving industry requires a sensor that performs at real time in all lighting and weather conditions. In addition, in a world where autonomous cars may drive one towards the other on highways at high speeds, the sensor must be able to “see” them coming from over 300 meters away, track velocity, and detect distance. In this presentation, Kobi Marenko will explain why Imaging Radar is the only technology that can overcome these challenges, and discuss what role Radar will have in an autonomously driven future.

Kobi will focus on how to increase public confidence in the autonomous market, while pushing the industry to develop further. He will specifically discuss the challenges that need to be addressed to achieve a next generation of Radars, such as sensing the road with both an ultra-high resolution and a wide field of view, resolving ambiguities, achieving low false alarm rates, coping with mutual radar interference, while keeping prices low and reliability high.
 

Intelligent Vehicle Data Analytics to Put Human in the Loop

User experience in the era of intelligent vehicles is shifting from the driver to the passenger. We can leverage from in-out cabin sensing/perception abilities of the intelligent vehicles and put human in-the loop and sense/perceive user in his various roles (driver, passenger, road user) in order to automatically adjust vehicle behavior to enhance user experience. In the new world of intelligent vehicles – we first need to address the needs of the  passenger/driver – and target  a natural and personalized in-cabin CAR HMI leveraging from the ability of the vehicle to sense and perceive the cabin and the passengers via inward facing sensors. Then we need to address the needs of the other humans – road users; ensure that new world of mobility  - a world where we humans are surrounded by a growing population of intelligent vehicles is a safer world where  traffic flows - advancing the quality of life.

 

Using Machine Vision and Machine Learning to Keep Passengers Safe and Comfortable

As a main goal of vehicles is to transport passengers, the well-being, safety and comfort of the passengers inside a car is of major importance. This becomes especially true when we envision the era of autonomous vehicles, and specifically autonomous public transportation systems such as autonomous taxis, in which there is no driver in the car to assume this responsibility.

At Guardian Optical Technologies we develop an in car multi-sensor to provide rich passenger data. Our sensor combines video images, depth data and micro-scale vibration sensing into one device. Based on these three layers of information we build different applications that can analyze and provide information on different aspects of what’s going on inside the vehicle cabin. The combination of these three different, complementary modalities give us great robustness and reliability.

As an example, a “forgotten infant” application monitors the inside of a locked vehicle to detect if any child (or pet) was left behind. We do this by utilizing the great sensitivity of the micro-scale vibration layer, capable of detecting breathing or even heart-beat motion even with no direct line of sight. Add with the depth and vision layers this result with unmatched detection robustness.

As another example, an “occupancy” detector is used to replace the current seat-belt reminder sensor. We use our sensor not only to detect that there is a person on the seat (and not just a heavy bag), but also to classify their mass and age, and to indicate if they are sitting correctly or are out of position, for smart airbag deployment in case of emergency.

The list of possible applications goes on. It starts with basic passengers and driver monitoring and ends with complex behavior analysis, such as restlessness or violence of passengers. We achieve this level of capabilities by employing state of the art machine learning and neural network algorithms, training the system with big data collection. In this talk we will demonstrate some of our applications and explain how we are able to obtain them.

Drive4U Locate, affordable precise and robust localization and mapping for Automated Driving

For highly and fully automated driving, a high precise global map is needed to complete and enhance the local perception to handle complex highway and urban use-cases including driving rules and context interpretation. To extract the relevant attributes from the map, decimeter positioning accuracy is required within this map.Commonly used solutions employ high precise GNSS positioning systems like DGPS, RTK which are too costly to be embedded in automotive grade products.To propose a cost effective solution to OEMs, Valeo developed the Drive4U Locate for precise and robust crowd-sourced simultaneous localization and mapping (SLAM). This solution enables centimetric accuracy using onboard Lidar sensors (Valeo ScaLa), affordable automotive grade IMU and low cost GNSS receiver. It works in poor and denied GPS conditions (tunnel, multipath issue...), it is suitable for indoor/outdoor parking and large scale areas and can be complementary to existing HD Maps.For the mapping, multiple vehicles collect and upload to the cloud a real-time online local map based on ScaLa data. A map fusion is done in the cloud to aggregate and update the map changes. This updated map can then be distributed to the subscribed automated vehicles. This allows these vehicles to locate themselves precisely using the updated map.This closed loop process guarantees the maintenance of a precise and up to date map for automated cars.Preliminary results from test drives with Drive4U Locate in urban roads in France and US show that centimetric accuracy can be obtained using current serial production Valeo automotive grade sensors.

Deep Learning Processing Technologies for Embedded Systems

Ushering in the Era of the Self-Healing Car

The era of software driven cars is upon us. With all the best intentions, the pressure to deliver car features quicker versus the rise in the amount of software in the vehicle means that not all the bugs can be found and fixed before the features ship. Current diagnostic tools only identify pre-programmed OBD-II errors and current OTA update solutions sit idly by waiting for new software versions to become available. Aurora Labs is ushering in the era of the Self-Healing Car. Using machine learning algorithms to uniquely address all three stages of an automotive software maintenance system, OEMs can now predict, fix, and seamlessly implement OTA updates to faults in the software. With the rising cost and frequency of software-driven recalls, a solution is required that enables reliable and cost-effective rollouts of new automotive features to all ECUs in the vehicle without any downtime for the user.

Next-Gen Computer Vision Cabin Sensing on the Path to Full Autonomy

These are exciting times for the automotive industry with the era of autonomous vehicles that is emerging upon us. Drivers now enjoy new levels of safer and better driving experiences thanks to groundbreaking Computer Vision and Artificial Intelligence that is paving way to a driverless future. With all the excitement around autonomous driving it certainly feels that human drivers will soon be a concept of the past, allowing us to rely on calculated machines and eradicate dangerous drunk, drowsy, and distracted drivers, but the reality is that we still have hurdles to pass until we get there. As we pave the path to full autonomy, advanced Computer Vision and AI technologies will play a pivotal role in the transition from low levels of autonomy upwards. The automotive landscape today is dominated by vehicles below level 3 autonomy, requiring an attentive driver behind the wheel, and when looking at road accident statistics reality is grim. In the U.S., auto deaths have jumped tremendously in the last three years with over 40,000 last year according to the NHTSA. Further, they attribute 90% of fatal accidents to human error, while claiming that 74% of car collisions occur due to driver inattentiveness in the 3 seconds prior to the collisions. AAA states that 21% of all fatal accidents involve drowsy driving. Computer Vision and AI solutions are key to battling these statistics by offering driver monitoring systems that provide an ongoing assessment of the driver’s state. Such systems track the driver’s eyelids, blink rate, head pose and even pupil dilatation to detect if the driver is drowsy, distracted or asleep, allowing to take proactive measures from audio alerts and vibrating seats, to reducing the speed or stopping on the shoulder. In level 3 autonomy, where there is transfer of control between car and human, driver monitoring technology accurately assesses the driver’s state to determine the right time for the handoff of control from car back to human.

Future sensors for autonomous driving: robust detection of any obstacle under any weather and lighting conditions

Ride Vision ARAS - 360° Threat Analysis using cameras ***VIEW PRESENTATION***

One Sensor To Rule Them All – Texas Instruments Imaging Radar

***VIEW PRESENTATION HERE***

This presentation describes the current ADAS/AV sensing technology market, 
survey the advantages and disadvantages of each sensor type in light of few of the accidents which recently occurred, and then introduce the Imaging Radar sensor, a new sensor technology, that we believe will be a must have sensor in level 3, 4 & 5 AV.
In the 2nd part of the presentation, we will go deeper and explain how cascade multiple TI single chip radars to a high performance radar sensor. The cascade radar system can support both MIMO and TX beam forming modes for high angle resolution and long detection range. The high accuracy phase shifter enables actively beam steering towards desired angle of interest. Some field test results are presented based on TI 4-chip cascade evaluation board to demonstrate the achieved performance

Deep Learning Autonomous Simulation

Hacking Automotive Ethernet Cameras

Hacking Automotive Ethernet Cameras
Daniel Rezvani, Argus Cyber Security

***VIEW PRESENTATION HERE***

Autonomous vehicles rely on a range of sensors to interact with the world around them. One of the most noticeable sensors is the Ethernet camera - a standard camera which is used for vision-based ADAS (advanced driver assistance system). Since this camera has become a critical part of an autonomous car’s safety (it is responsible for identifying nearby hazards, traffic signs, etc.), the consequences of a successfull cyber-attack launched against such a camera can be devastating, resulting in real physical injury or death. In this technical presentation, you will learn how our team of automotive cyber security researchers were able to easily hack an Ethernet camera (similar to those being integrated into today's connected vehicles) and trick it into thinking the pre-recorded video is reality.

 

GuardKnox

3 Pillars of Autonomous Driving

***VIEW THE PRESENTATION HERE***

Autonomous Driving 2.0

ICT for the Future Mobility

Individual transportation is hit by three mega trends, which are electrification, automation, and the digital transformation. All these three mega trends are also vital questions for the traditional car manufacturer and therefore of tremendous impact on the automotive eco-system existing for more than 100 years. Electrification simplifies the drive train enabling new player in the automotive market on the one side of competition while automation will place a taxi ride into a fully new competitive position against private car ownership. In consequence car sharing offerings in major economies will be boosted, and in turn sharing customers will be open for new modalities like the electrical vertical take-off and landing (eVTOL) vehicles being on the horizon as a convenient alternative for medium distances. Despite both approaches, privately owned cars from traditional OEMs and any car sharing fleet, will become electric and autonomous, the technological approach might be different. Traditional OEMs are used to offer their customers flexible individual all-in-one vehicles and therefore developing a corresponding stand-alone autonomous driving experience. In contrast car sharing operators are focused on return on invest within a fleet approach. Consequently fleet automation will follow other paradigms in comparison to car automation and will therefore drive other technological solutions. A supporting road and back end infrastructure in case of fleet automation incl. flying vehicles, decoupled from energy consumption constrains of the BEV electronics, drives potentially other solutions as currently developed within car industry. Fleet operation will especially benefit from the digital transformation driving administrative expenditures to a very low end.

A draft scenario of these developments is going to be presented as a discussion base within the community to develop a joint understanding of future technology demand. Will these developments drive competition or will the community see a cooperative environment? We intend to initiate a discussion on technology development for the future automobile eco-system.

The role of safety, robustness, and explainability of AI in the physical world; a Tier-1 perspective.

Artificial Intelligence (AI) has been a story of great success over the last years in many fields and application domains. Technologies such as machine learning or planning - often grouped under the general term AI – are substantially contributing to make all kind of products more intelligent and human friendly, as we will illustrate on several Bosch products. It is widely accepted that AI technologies are also key to autonomous driving and the Internet of Things. Specifically in recent years, the focus in the automotive space was about perception and in-vehicle UX. However, as AI technologies are increasingly applied in the physical world, ‘under the hood’ and in the production line, new levels of rigor and guarantees are required. We will discuss the role of safety, robustness and explainability of AI, showing where research, which contributes to develop highly reliable standards for AI, is needed in the engineering domain

Implications of the Software Super Inflation

In the past years, cars are becoming more and more dependent on electronics and software. Modern cars have between 50 and 100 Electronic Control Units (ECUs), from smaller micro-controllers to powerful computers running open operating systems such as Android. These cars also have up to 100 million lines of code. As a comparison, this is more than 10 times the number of lines of code in NASA’s space shuttles, and 5 times more than the new Boing 787 Dream Liner airplane. It is estimated that software today accounts for almost 10% of the car value (BOM).
When we look 5 to 10 years ahead, these numbers are expected to grow significantly, with the introduction of ADAS and self-driving features, digital cockpits, cloud services, heads up display and more. Estimates indicate somewhere between double and triple the amount of lines of code, to between 200 and 300 million lines of code, and up to 30% of the BOM.
The implications of this super inflation of software in future cars includes the obvious fear of bugs and ultra-costly recalls that may result, but also implications on design, production, supply chain, service and more. Car makers (OEMs) already started to get themselves prepared, mainly with the introduction of Remote Software Update solution, but they will have to think deeper on several other changes they will have to make to be ready.
In his presentation, Oren will review multiple implications of the software inflation, and the different aspects OEMs need to think about when addressing the issue, and preparing for the implications.
Oren is leading HARMAN’s Remote Software Update business, who was already selected by 20 OEMs, representing almost 300 million connected vehicles in the coming 10 years.
About HARMAN
With a workforce of some 30,000 automotive professionals across the Americas, Europe, and Asia, HARMAN designs and engineers connected products and solutions for automakers, consumers, and enterprises worldwide. HARMAN’s broad-array of cutting-edge products includes digital cockpit systems, Remote Vehicle Updating over-the-air (OTA) solution, audio products, SHIELD Cybersecurity Solution and IoT services. In March 2017, HARMAN became a wholly owned subsidiary of Samsung Electronics Ltd.
Audiophiles from every generation call on HARMAN to deliver the best in sound in the studio and on the stage, in the car, at home and on the go. HARMAN’s portfolio of legendary audio brands includes AKG®, Harman Kardon®, Infinity®, JBL®, Lexicon®, Mark Levinson® and Revel®.
Since 2012, HARMAN has made significant and ongoing investments in Israeli technology, including the acquisitions of iOnRoad, Redbend and TowerSec. HARMAN Automotive Cybersecurity is a full service global business unit that has deep experience in traditional IT security and embedded security, as well as several years of pioneering work in automotive cyber security. HARMAN’s Remote Vehicle Updating offering is the only OTA solution that enables efficient full-vehicle management, already chosen by 20 OEMs to enable secure OTA software updates to more than 30 million vehicles.
In November 2017, HARMAN has announced the launch of the International Cyber Security Smart Mobility Analysis and Research Test (SMART) Range in Israel, in cooperation with Ben-Gurion University of the Negev. HARMAN has been an active member of industry institutions - such as Auto-ISAC, SAE/ISO, Jaspar and others – working to enhance cybersecurity awareness, design and deployment across the global automotive industry.
 

The Road to Autonomous Driving: The Challenges and Opportunities of In-Vehicle Connectivity

• The need for increased bandwidth, as more and more cameras, sensors, and telematics are added
• The need for a reliable, light-weight and easy to use cable for transmitting data
• The need for robust and adaptable connectivity to handle the rough automotive environment
• The need for a scalable solution

Continuous Deep Learning at the Edge

***VIEW PRESENTATION HERE***

The robustness of end-to-end driving policy models depends on having access to the largest possible training dataset, exposing the true diversity of the 10 trillion miles that humans drive every year in the real world. However, current approaches are limited to models trained using homogenous data from a small number of vehicles running in controlled environments or in simulation, which fail to perform adequately in real-world dangerous corner cases. Safe driving requires continuously resolving a long tail of those corner cases. The only possible way to train a robust driving policy model is therefore to continuously capture as many of these cases as possible. The capture of driving data is unfortunately constrained by the reduced compute capabilities of the devices running at the edge and the limited network connectivity to the cloud, making the task of building robust end-to-end driving policies very complex.

Bruno Fernandez-Ruiz offers an overview of a network of connected devices deployed at the edge running deep learning models that continuously capture, select, and transfer to the cloud “interesting” monocular camera observations, vehicle motion, and driver actions. The collected data is used to train an end-to-end vehicle driving policy, which also guarantees that the information gain of the learned model is monotonically increasing, effectively becoming progressively more selective of the data captured by the edge devices as it walks down the tail of corner cases.

Autonomous Driving in Unknown Areas

Autonomous driving has come a long way albeit it's success is limited only to those areas where a High-Definition map is pre-built and known. Inspired by the recent achievements in Artificial Intelligence particularly of methods that combine trees and DNNs (e.g. Alpha Go Zero), this talk demonstrates how to effectively combine Deep Learning and convention path planning to drive in unknown areas.

How safe are autonomous cars and what are the industry doing to make them safer – Functional Safety

Functional Safety is all about ensuring the safety of operation of electronic devices, in all conditions, even when hardware failure happen. With the rise of autonomous driving, Functional Safety is gaining more and more importance and its becoming key requirements in all automotive electronic systems. In this presentation, we introduce the standard ISO-26262 that is enforced by car OEMs to guarantee the Functional Safety of all devices they put in the car, how the standard is used to communicate requirements between OEMs, Tier-1’s and Tielr-2’s, we touch on its requirement for system-level, software-level, and electronic chips level.

End to End Vehicle Lateral and Longitudinal Control

In this presentation, we will present a complete study of an end-to-end imitation learning system for lateral and speed control of a real car. Indeed, as autonomous vehicles are accumulating mileage, there are two main research streams to deal with increasingly difficult situations, in particular urban driving, with a high amount of reliability to get increasing levels of autonomy.
In the most common approach, the autonomous driving functionality is cut into different modules, from perception to control, going through data fusion, path planning etc. An alternate method is to consider the vehicle behavior in an end-to-end fashion: the raw low level data from the sensors (classically cameras) is more or less directly used to produce the car commands. This is described as “behavior reflex”, because it is a reaction to the environment without explicit description. Using alternate approaches such as end-to-end, even for limited tasks, can be combined with more classic pipelines, in order to provide new redundancy channels that are mandatory for safety purposes.
To train a neural network to produce end-to-end controls, one can let the network learn by itself the way the vehicle should behave by trial and error: this is Reinforcement Learning. Another way of training would be to take the ground truth from an external behavior (typically, a human driver), and train the network to produce the same control output using the raw signal. In literature, this is called behavioral cloning or imitation learning. In this study, we will describe how to design an end-to-end imitation learning system for Lateral and speed control of a real car, learn two networks, one for lateral control and one for longitudinal.
Concerning networks to control steering angle for autonomous cars, most of research publications use multiple long range cameras to generate failure lateral cases in order to handle error accumulation issue when testing on real car. In this presentation, we present a novel model to generate this data augmentation using only one short range fisheye camera. We then present how we build a simulator and how we used it as a consistent metric for lateral prediction evaluation. Experiments are conducted on a new dataset corresponding to more than 10000 km and 200 hours of open road driving. We evaluate this model on real world driving scenarios, open road and a custom test track with challenging obstacle avoidance and hard turns. The final model was capable of more than 99% autonomy on urban road.
Concerning lateral control, we propose a neural network based on an LSTM and propose to use a loss that integrates dynamic information in order to let the system learn speed control dynamic similar to the one produced by a human driver. We also propose data augmentation and label augmentations that are relevant for imitation learning in longitudinal control context. Based on front camera image only, our system is able to correctly control the speed of a car. More precisely, the system is able to control the speed of a car in simulation environment, and in a real car on a challenging test track. The system also shows promising results in open road context.
Finally, we will describe how we have integrated the two networks on Valeo research prototype cars, used as platforms, both for collecting data and for testing. Cars were equipped with an active drive by-wire control, which made it possible to send steering wheel angle and acceleration commands to the car. To run the network inference on board, an embedded GPU board was used. Our experiments show that the proposed system is able to drive the lateral and control car speed in both simulation and real cars. We will present results of these experiments and show videos of open road test and demonstrations done at CES 2018.

Challenges in Implementing IoT in Your Products and Building Your Organization Go-To IoT Strategy

Mobile Livingroom 2.0

Most expect the car to be the third place of living aside from our home and our office. Longer distances to commute, more traffic, more time in the car is creating this reality already today. But how can we avoid that this development turns out to be our personal mobile dystopia? How do we create a place to be and a place where you like to be? HMI, AR, Interior Cocoon and new perceptual safety features are core to create a Mobile Livingroom 2.0. - These concepts will be introduced, explained and discussed.

Accidents, Scenarios, Verification and the path to safer autonomous vehicles.

That tragic Uber accident has brought Autonomous Vehicle (AV) safety into sharp focus. It raised awareness of aspects like how people are more afraid of things they can’t control, the need for third-party testing, the insurance implications of all this and so on.
Regardless of the specifics of this incident, this presentation will look at the bigger picture of AV deployment, safety and verification, and expand on the following claims:

Beyond a certain safety threshold, AVs should be deployed
While safer than human drivers, AVs will continue to have many fatal accidents
AV manufacturers and regulators should employ a well-thought-out, comprehensive, continuously-improving, multi-execution-platform, transparent verification system

There seem to be a need for the various stakeholders (the public, lawmakers, regulators, AV manufacturers etc.) to agree on some general framework for handling these accidents (and the whole deployment process). That framework should ensure, among other things, that:

Not every accident results in a lengthy, billion-dollar lawsuit
Negligent AV manufacturers do get punished
Everybody (the public, the press, judges, lawmakers, regulators etc.) has an understandable way to scrutinize the safety of various AVs, both in general and as it relates to a specific accident scenario

This is going to be a non-trivial framework: It will surely have legal and regulatory components. It will probably include ISO-style “process” standards, such as ISO 26262, the SOTIF follow-on, and the expected “SOTIF-for-AVs” follow-on to that. It may contain a formal component and more.
But the central component (tying all others together) is probably going to be a verification system. The presentation will describe how an AV verification system lets you:

Define a comprehensive, continuously-updated library of parameterized scenarios
Run variations of each scenario many times against the AV-in-question, using a proper mix of execution platforms (such as simulation, test tracks etc.)
Evaluate the aggregate of all these scenarios / runs (and any requested subset of it), to transparently understand what was verified (this is called “coverage”) and what “grade” it got

Such a coverage-driven verification system (enhanced by ML-based techniques) is probably our best bet. It will also be a crucial component for improving safety as quickly as
The presentation describes the main attributes of such a system. A Scenario Description Language will be described, and a tools suite to utilize it and create   a path to safer autonomous vehicles is described and presented.   Specific focus is given to regulatory aspects of such a language, and its potential usage as a certification tool.
 

Turning an Automated System into an Autonomous system using Model-Based Design with MATLAB

An autonomous system is a system that functions independently or in a supervised manner and operates under conditions of uncertainty, in an unknown and unpredictable dynamic environment.

Autonomous systems may accumulate new knowledge or adapt themselves to a changing environment. In order to complete their mission, these systems can acquire information from their surroundings, move independently in their environment (real or virtual), avoid dangerous situations and act for a long period time without human intervention.

MathWorks Model-Based Design (MBD) approach, which is based on MATLAB & Simulink, includes many capabilities that allow us to design, simulate and test autonomous systems in a simple and comfortable workflow.

 

This lecture will review a number of capabilities from the autonomous domain. These capabilities will allow us to transform the designed system into an autonomous system that can make decisions independently. During this session, we will present several possibilities for designing an automated system, define and combine different sensors (Sensor Fusion), create perception capabilities in order for the system to better understand its environment, planning and following optimal trajectories, and finally, how our autonomous system can interface with other environments.

This MBD workflow allows us to save a significant amount of time during the development process, up to the prototype stage and on.

Driving the future of connected vehicles

***VIEW THE PRESENTATION HERE ***

This presentation will discuss what will happen on our roads until we reach the stage (if at all) when all vehicles will be autonomous. It will present how and why autonomous and manned vehicles must find a way to share our roadways.
Questions will be raised regarding whether or not autonomous vehicles can predict human behaviour and vice-versa, can human drivers anticipate the intentions of autonomous vehicles.
 
Other issues will be raised regarding how to protect vulnerable road users and whether “extra” protection will be needed for motorcycles, bicycles and pedestrians.
V2X will be presented as a solution and additional benefits of this technology will be highlighted including improved safety, mobility, and emissions.
A conclusion will be presented discussing all the main reasons why V2X is an essential technology on the way to full vehicle autonomy.

 

How deep learning is enabling unprecedented driver and in-cabin monitoring using cameras

The topic of driver monitoring is 20 years old by now. With the advancement of autonomous vehicles, interior monitoring is gaining importance:
- For L2 driver monitoring to increase safety by identifying drowsiness, distraction, and personalization of the cabin
- For L3 driver monitoring to enable the handoff control between the semi-autonomous vehicle and the driver
- For L4+ in-cabin monitoring is critical as there is no driver to watch around, e.g to know how many people came in, if they have their seatbelts or not, any medical or distress conditions and so forth
Recent breakthroughs in deep learnign and AI enable us now to do automotive grade driver and cabin monitoring, from various cameras and positions, and running on existing compute inside the vehicle, giving OEMs the critical view of the driver and cabin they need.
In the presentation we will show some of Jungo`s breakthroughs in AIs, and the use cases we enable for upcoming upcoming cars, already licensed by multiple OEMs and Tier 1`s.