Seriously Real Time Data Processing

  • On the self-driving car thing, it weould be much easier if all cars were self-driving. Take the human idiot factor out of it and it instantly becomes simpler, especially in a motorway situation where traffic is all flowing in the same direction and there are slip-roads rather than intersections.

    It is obviously much more difficult in urban situations with pedestrians, cyclists, etc. which is the huge challenge.

  • Real-time? This depends upon who is talking.

    Having worked in this sector, e.g. where I have produced reports for people on demand, and in real-time microprocessor-driven system, I can say that the what is meant by "Real Time" in the two fields is quite different.

    Even though the report may come back quickly, after the user has requested it, the the times for this are neither consistent (varying dependent on a range of factors) or controllable by the user. I would very much not liked one of my real-time microprocessor-driven systems to behave like that! However, in the It-Systems world, if the user asks for a report and it comes back within 4s (I was told this was a target in an SQL Server Central article a long time ago) then it is 'Real Time'.

    For example, in a one of my control systems, it would have to initiate the inputs at time 'A' (normally analogue-to-digital conversion), process the results when they arrive and produce an output (to a control circuit) at time 'B'. The time 'A' to 'B' being, for example,  2144µs not 2143µs or 2145µs or anything else. The processing is fully under control and no variation is allowed. This is because it has to work in the real world, like the self-driving cars. Another way of looking at it is that it has to synchronise with physical reality whereas the production of the reports is asynchronous - they arrive when they arrive - not 'Real-Time' at all according to control systems.

    • This reply was modified 4 months, 2 weeks ago by  nicksamuel.
  • One thing engineers can do is install the sensors and self driving software in cars - but just don't connect it to the steering, breaks, etc. Human drivers, who volunteer in the project, can go about their usual daily driving routine, and then engineers can review the telemetry and recorded video to see how the software would have responded to various conditions and events.

    For example: Would the software have responded sooner than the human to that dog who darted into the road? Would the software have chosen to change lanes, and was the alternate lane clear? Did the software recognize that down powerline or brick in the road as a hazard?

    All this can be simulated in a controlled environment, but there are some circumstances that might crop up during millions of hours of diverse driving that the engineers might not think of.

    "Do not seek to follow in the footsteps of the wise. Instead, seek what they sought." - Matsuo Basho

  • Chris M wrote:

    On the self-driving car thing, it would be much easier if all cars were self-driving. Take the human idiot factor out of it and it instantly becomes simpler, especially in a motorway situation where traffic is all flowing in the same direction and there are slip-roads rather than intersections.

    It is obviously much more difficult in urban situations with pedestrians, cyclists, etc. which is the huge challenge.

    I'm surprised this hasn't been done more. There are a few small test areas, but I' expect that somewhere in a big city, this would be come reality, with taxis and such being the only thing moving people around.

    At this point, I wonder if we'll see this first on highways, like an HOV type lane, where only automated vehicles drive.

  • Eric M Russell wrote:

    One thing engineers can do is install the sensors and self driving software in cars - but just don't connect it to the steering, breaks, etc. Human drivers, who volunteer in the project, can go about their usual daily driving routine, and then engineers can review the telemetry and recorded video to see how the software would have responded to various conditions and events.

    For example: Would the software have responded sooner than the human to that dog who darted into the road? Would the software have chosen to change lanes, and was the alternate lane clear? Did the software recognize that down powerline or brick in the road as a hazard?

    All this can be simulated in a controlled environment, but there are some circumstances that might crop up during millions of hours of diverse driving that the engineers might not think of.

    I assume some of this is done in testing already. At least, I hope they're doing some of this.

  • Chris M wrote:

    On the self-driving car thing, it weould be much easier if all cars were self-driving. Take the human idiot factor out of it and it instantly becomes simpler, especially in a motorway situation where traffic is all flowing in the same direction and there are slip-roads rather than intersections.

    It is obviously much more difficult in urban situations with pedestrians, cyclists, etc. which is the huge challenge.

    This is almost verbatim what I was going to comment. The biggest chaos monkey in the system is the loose nut behind the wheel. Followed by strange/rare events the engineers couldn't predict.

  • For a sceptical and very knowledgeable view of the current state of AI and robotics I can recommend Rodney Brooks' blog

    https://rodneybrooks.com/predictions-scorecard-2021-january-01/

    Brooks was the director of the AI lab at MIT until 2007 and currently runs an industrial robotics company.

Viewing 8 posts - 1 through 8 (of 8 total)

You must be logged in to reply to this topic. Login to reply