rr6013
Well-known member
- First Name
- Rex
- Joined
- Apr 22, 2020
- Threads
- 54
- Messages
- 1,680
- Reaction score
- 1,620
- Location
- Coronado Bay Panama
- Website
- shorttakes.substack.com
- Vehicles
- 1997 Tahoe 2 door 4x4
- Occupation
- Retired software developer and heavy commercial design builder
- Thread starter
- #61
DOJO heuristics abstract and simplify for FSD this human visualization pictured. Remember this diagram is for human consumption. What FSD will see is entirely different. The pretty picture tells nothing about the chain of AI decisions which lead to SoC commands that affected the behavior of the mechanicals of the vehicle to inflict accident or injury AFTER DOJO heuristics.The FSDv9 visualization is already "explaining" what the AI sees and interprets.
.
It took the 5th most powerful supercomputing machinery in the world optimized for NN/ML to analyze the data to create FSDās heuristics in this first place. FSD canāt analyze itself anymore than a human.
I have doubts diagrams, logs and register stacks are useful beyond establishing āstateā of FSD. I watch AP Fail video and read about in-kind failure with emergency vehicles and it simply appears that AP is blind. Why, how long, is it timing, throughput bottleneck, architecture, code constraint? The glaring blindspot for Tesla is tractor-trailer stretched across lanes of travel. Still no progress stopping those accidents eliminating that 90Ā° blindspot.
So this is a criticality to define the situations in which it must be immune to Depth of Vision attack by pulling across lanes of travel, oversize billboard on a curved road and topāo grade skyviews that could flash FSD blind(lightning, sunset)
Situational awareness attacks in the 90Ā° plane have been fatal flaws for Tesla self-driving. Its begun to fall into the class of architectural flaw since its not structural component level fails.
Sponsored