Keeping The GoPiGo3 Platform "Current"

[cyclicalobsessive]

@cyclicalobsessive

… Bookworm Raspberry Pi OS … for GoPiGo3?

• 32-bit I would not touch with a 10 foot pole.

@jimrh

Why not?

I think of 32-bit anything as “compatibility for old stuff”, in the case of GoPiGo3 - really old stuff now.

PiOS Bookworm is the latest evolution for the Raspberry Pi single board computer family that includes one processor that slightly broke compatibility (Pi4) and one processor that placed a land mine in compatibility (Pi5). Running 32-bit Bookworm for compatibility with unmaintained software written before the Pi4 era is an oxymoron to me.

Software compatibility and computing platforms has always been a moving target, but the last 10 years has given me whiplash looking at all the new platforms and new features on existing platforms. The next 10 years is going to “legacy-ize” any platform without a massively parallel processor (GPU/AI processing unit).

The key to using vision on a mobile platform at reasonable response/frame rates is a multi-core general purpose processor with a dedicated parallel processor. We just saw announcement of an over-$1000 single board computer:

Particle has launched Tachyon , its first Qualcomm Snapdragon-powered single-board computer (SBC) designed to make cutting-edge chipsets and AI tooling widely accessible to consumers and businesses.

To bring GoPi5Go-Dave into the current generation of mobile robots, I had to figure out a way to run the unmaintained 32-bit era GoPiGo3 API on the latest Raspberry Pi, and add a camera with a robotics vision processor:

• 2xLeon CPU cores: USB/ethernet, IMU, 3A algorithms. scheduling
• ISP - Image Signal Processor - denoising, sharpening, etc.
• 2x NCEs (Neural Compute Engines)
• 16x SHAVE cores - vector processors - NN operations/layers, CV algorithms
• 20x CMX slices - fast SRAM memory blocks (each 128kB)
• Stereo pipeline - Stereo matching (census transform, cost matching and cost aggregation)
• Video encoder which supports MJPEG, H264 and H265 codecs.
• Vision blocks:
  • Edge detection
  • 3x Warp engine
  • Corner detection (Harris/Shi-Thomasi)
  • Motion estimator
  • Min/Max calculator - FeatureTracker, corner detection

The GoPiGo3 just missed inclusion in a 2015 article Can Robots Reshape K-12 STEM Education?

(quoted over 235 times by researchers and included 36 commercially available robots available when the GoPiGo3 was being introduced.

It takes big, long investment to create something that can be both a profit vehicle and an educational vehicle.

• FIRST (For Inspiration and Recognition of Science and Technology) has over 200 corporate sponsors (including Qualcomm proudly on the FIRST landing page ) and is probably the most successful at integrating robots into the schools.
• VEX Robotics may be the only single corporation to have succeeded at integrating their robots into schools at scale.

Oh wow, I guess I managed more than 20 character minimum on this post!

[cyclicalobsessive]

@cyclicalobsessive

the latest evolution for the Raspberry Pi single board computer family

Actually, the latest evolution announced today is the Raspberry Pi Pico 2. Talk about evolution - looks like a step toward a plan to obsolete the ARM processor and move to RISC processors to cut cost.

[jimrh]

@cyclicalobsessive

looks like a step toward a plan to obsolete the ARM processor and move to RISC processors to cut cost.

It’s not necessarily just cost cutting.  It’s very likely that people are beginning to finally face the reality that CISC processors have reached a technological dead-end.

“Back in the day”, HP used RISC processors on its servers and the DEC Alpha was an attempt to make a RISC-like processor fit DEC’s computer landscape.  The Itanium processor was Intel’s attempt to do the same thing - fit the square peg of a RISC-like system into the round hole of Intel’s CISC processing model.

RISC had serious advantages back then.  It was “cheaper, faster, better” than a comparable CISC processor.  CISC computers have closed the gap because they crammed more and more gates onto the die and ramped up the processor speeds.

RISC wasn’t popular, (and didn’t really “catch on”), because it represented as huge a paradigm shift in the processor model as the original Mac, (along with its desktop and windows), was in the user interface world - and people back then couldn’t bridge the gap with the primitive tools and the linear, single threaded processing model thinking that everyone was taught.

It’s like a RPN calculator in a “precedence rule” calculator world - it takes some getting used to, but once you get the hang of it, you’ll wonder how you ever got along without it.

Unfortunately, Moore’s Law has had to face the realities of Heisenberg’s Uncertainty Principle and the constraints our current technology place on die density:  Only so many electrons can dance on the head of a (nano) pin.

IMHO, without a “quantum leap” in our understanding of nano and pico level physics, we’ve reached the practical limit of our ability to make CISC processors.

Also IMHO, until quantum computing becomes mainstream, (if ever), the future will depend on massively parallel computing with hundreds, maybe even thousands, of simple RISC processing pipelines, similar to the NVIDIA Cuda processors today.