Get email delivery of the Cadence blog featured here
DVCon is one of the premier conferences WW for design and verification. The DVCon India show has grown significantly over the last 4 years of its existence and this year’s edition was as vibrant as previous years.
Cadence’s Apurva Kalia gave a keynote in the Design and Verification track at DVCon India 2017. The talk had the catchy title, “Would You Let Your Child Ride In An Autonomous Car?”, which I thought was a nice title for this blog post.
Apurva started with tracing the evolution of the car – from the wheel, to horse power, to the first car 100 years ago. And within 100 years we are talking about autonomous cars - that’s quite a feat of innovation!
He noted that that every few years there is a cycle in the semiconductor industry when a disruptive technology emerges – it pushes the technology envelope and drives the fortunes of semiconductor companies, fabs and other ecosystem players. The last one was mobile computing, which is leveling off now. Data shows that the next big cycle could be electronics in automotive. ADAS is driving growth and is growing at a 22% annual growth rate.
Saying that making autonomous driving a reality is a hugely complex task is stating the obvious. What are some of the challenges?
The first challenge is to make the car “aware”, and this requires complex sensing. Ultrasonic, radar, LIDAR (laser-based radar sensor) sensor, standard camera based techniques – each of these technologies have pros and cons and they all have to overcome those to work together to provide an accurate sensory system. A fusion of sensors, so to speak.
The next challenge is perception. How does the machine make sense of all this data coming in? Apurva said that in machine learning, you do not tell the program what to do. The machine has to discover by itself based on previous learning and experience. With the environment being so complex, this challenge is very real. The data is complex and at any given time the system has to keep track of multiple objects, each with a different size, moving in different trajectories and speeds, and in a different manner – think of cars, pedestrians, cyclists, even cows!
The good news is that while on one hand the environment is complex, the training data is available. All you have to do is drive around with a camera on your dashboard, and the camera is capturing all the data that can be used for training machines. Google has been doing this for years, they have got millions of hours of training data from various cities around the world.
Many car manufacturers have already included some amount of ADAS in their recent models. Lane departure warning, forward collision warning, lane keeping assist – these are just a few of the features that are becoming common, and that too at not much additional cost.
Coming to an Indian example, Apurva talked about an innovation by students of a premier Delhi engineering college - a self-driven 3 wheeler! He said that he was at a presentation where they showed a video of it successfully navigating through the congested Govindpuri area in Delhi – quite a notable achievement!
Even with a little imagination you can grasp the horrors of accidents with autonomously driven vehicles, where one may be helpless to intervene.
Apurva said that according to the SAE International’s standard J3016 that defines six levels of automation, we are at level 3 right now (“Conditional Automation” which he interpreted as "hands-off"). In active development is level 4, which is “High Automation” or “eyes off” – it is in prototype stage. The final stage 5 is “Full Automation”, or what Apurva called “mind off”.
From a safety perspective, certification is critical when designing for autonomous vehicles, considering the significant safety concerns. Functional safety principles and certifications like ISO 26262 and ASIL certifications are important. Most ADAS systems are targeted at D-level of certification. Apurva also said that from a design perspective, we need a new methodology to ensure the vehicle's safety. As he illustrated through his presentation, today existing technologies are coming together in a new way. Not only do they have to work together in a flow or methodology, but that methodology has to work with existing methodologies.
So, in summary, to answer the question posed in the title: Apurva said, that yes, we will soon be able to send our children to school in a self-driving bus.