User and Data stories for a Practical Autonomous Vehicles System

— Part 2

Its an appropriate time to recognize that autonomous driving system decisions and extent of control is still in flux in terms of what behaviours would define full autonomy ?

Understanding this flux means reflecting on the fact that not only regulations but also demographics, talent pool, tech affordability, local environment and education will drive deployment and adoption of different levels of autonomy.

What I meant by that is for e.g. the operations and extent of control in tricky situations by a level 5 AV vehicle is pretty much subjective to creativity, standards if any and above all factors.

I would also like us to think about the extent of human driver accompaniment and involvement corresponding to a particular level of autonomy. With those things in mind, following is the gist of what ought to be considered in fully autonomous driving systems:

Safety, standards, learning, driver access control and detection of threat for autonomous vehicle system.

1. What is the Purpose of the autonomy ? reduce the incidents, make roads more safe, take some of the driving stress out for commercial transporters and commuters, optimize the use of infrastructure and energy.
2. Do companies think about the purpose when making everyday design choices in autonomous vehicle technology and infrastructure ? For e.g. would be good to push for the use of data by doing an in depth analysis of major fatalities, causal analysis, simulate failure situations and aim for training and perfecting full fault tolerant, smart navigation for at least 10% of those incidences with a timely target? That would be proving ground for 10X gain from the state of the art( with human drivers).
3. Driving away the intelligence hype — Do humans really use only intelligence in driving? The act of getting from A to B is a mix of

common sense, perception,civic values, training/practice and ultimate need of judgement/reflexes (not just intelligence).

(There is some empathy, some courtesy when we talk about civic values. Is our vehicle tech designed to simulate, input or measure all of these and use/ simulate and augment the decision metrics ?)

4. Eliminating factors of human constraints, yet not aiming for superpower autonomy — Few human constraints such as

1. Wakeful hours
2. Hearing ranges
3. Attention span
4. perception rate and planning horizon
5. Habits
6. Most importantly, mental state and focus.

When building scenarios, thinking critically of which one of above constraints, if solved, would practically be useful ? is a key to what extent we want to push to. Maybe we don’t have to try that hard (ultimately human definition of comfort would limit the use of perfect the system anyways, aka. users may not be comfortable with the superpowers, hence don’t make it smarter than needed).

5. Tricky situations and conflicts of interests - Human driver and passenger vehicle interactions — would both passenger and driver be in agreement for driving decision ? What is the ethical boundary of system deciding who gets to control the commands and the vehicle itself. What if one of them changes its intentions? Will the vehicle be able to determine, interrogate and overcome such situations with its own good intentions ?

6. Enforcements — Why would we need the algorithmic enforcements? — How about factors like user programmable mode for autonomy(defining default path and behaviours, should there be an emergency), no ignore/by-pass options should be possible for so and so road conditions. No overrides and vehicle is able to follow law from first order bit to last one. Behaviour enforcement for driver in the seat — Additional measures to detect things like driver’s fingerprint, weight, posture, face detection and biometric measures to make sure that the vehicle system is not getting compromised by intruders would be a good safety measure.

7. Standardization of operations — Thinking about standard implementation of autonomous vehicle commands and controls so that a person used to operate BMW’s autonomous car can, when needed can operate Volvo’s self driving car as well.

In practice, system behaviour, constraints, representation of states, vehicle negotiations and command and control all would be standardized, then carefully designed goals around the purpose of autonomy would be rigorously tested. We would need different design and productionalising paradigms and have a way to challenge the system so support of the dynamics of decision making in autonomous vehicle by autonomous vehicle.

In the next post we will go over examples of how to achieve these and will write user stories shifting the focus from driver being user of the tech to the vehicle being user of the technology, sensors and software. This shift of thinking would prove to be very fundamental to the operations we device to support autonomous driving.

To be continued…into Part3