3 ms·

In the long-term, I think blockchains will make a major showing in the design of self-driving vehicles (SDVs). For SDVs to really take off, it needs to surpass

by ayewo 6y ago

In the long-term, I think blockchains will make a major showing in the design of self-driving vehicles (SDVs). For SDVs to really take off, it needs to surpass the safety of a human driver but the only way to do this reliably in a way that avoids the high costs and ugly protrusion of roof-mounted LIDARs is for SDVs to use low-cost internal sensor inputs and fallback on low-cost external sensor outputs from nearby vehicles[§] and road infrastructure. Essentially, what I think is missing from all SDVs currently in development today is an important ingredient behind why Internet-use is so widespread today: a common communication protocol between SDVs from different manufacturers. Once a common protocol is adopted/legislated, all SDVs will be expected to record any accidents encountered during any trips in a partial copy of the “road blockchain”. Stated differently, details of accidents and near misses will be recorded with timestamps on the partial blockchain maintained by each SDV in a way that the exact position of all SDVs within the vicinity of a car crash can easily be reconstructed. Of course this will mean SDVs will use clocks that are network synced. If 6 cars are in the vicinity of a crash between 2 cars, the trajectory of all cars before and after the crash could be reconstructed using the blockchain data from those 2 cars. If the 2 cars’ “black box” are lost due to them exploding on impact, the data could still be reconstructed using the partial blockchain data from the 6 SDVs proximate to the crash. The “road blockchain” would be especially useful in accident investigations but its privacy implications would need to be properly thought out. §: If a SDV has front sensors that start to return conflicting inputs while in operation due to inclement weather for instance, rather than the onboard computer aborting the ride in the middle of nowhere, it could switch to operating at partial instead of full situational awareness. In this state it could routinely query any three nearby vehicles for help: the vehicle in front, behind it & to its left to regain some level of situational awareness. Another use case would be falling back on wireless beacons when Internet access over cellular is lost. The SDV could start an adhoc network over wireless where it can broadcast a request for help from nearby SDVs. Something like: “I’m traveling from LA-SF. I need a recent copy of the point cloud from anyone traveling in the opposite direction (SF-LA)”. For this to work at scale, a common protocol that allows a SDV to start a wireless conversation with any nearby SDVs would be required.

geofft 6y ago
You need a mechanism to make sure that a particular vehicle is being trustworthy about what it adds to the blockchain. A malicious entity could just report that other vehicles were at fault. One straightforward way to solve that is to say that only known car manufacturers may upload results and it must be signed by something that chains to the manufacturer's key - but at that point, you don't need a blockchain for this.
- ayewo 6y ago
  If you restrict to manufacturers, what about people who modify their vehicles? No amount of signing will be foolproof—tinkerers will find a way to extract signing keys. The SDV-to-SDV protocol could borrow from the contact tracing protocol used for COVID which requires about 60% of prior contacts for a positive match [§]. If SDV A says that at 11:00pm today it passed SDV B, C, D at location X, then if we were to reconstruct the blockchain from the perspective of each of SDV B, C or D, the collected data should be identical to the chain reported by SDV A, otherwise we would simply discard records that conflict with the longest chain, similar to how the blockchain mechanism works with Bitcoin. §: https://ncase.me/contact-tracing/ https://ncase.me/contact-tracing/
michaelt 6y ago
If a self-driving car needs data from other self-driving cars on the road, how's it going to work in the early stages, when only 0.1% of cars on the road have the feature? IMHO it'd be much easier to sell people a $10,000 self-driving option that worked immediately, than a $2,000 self-driving option and a promise it would start working at some point in the future, when xx% of vehicles on the road had deployed it.

by ayewo 6y ago

geofft 6y ago

You need a mechanism to make sure that a particular vehicle is being trustworthy about what it adds to the blockchain. A malicious entity could just report that other vehicles were at fault. One straightforward way to solve that is to say that only known car manufacturers may upload results and it must be signed by something that chains to the manufacturer's key - but at that point, you don't need a blockchain for this.

ayewo 6y ago
If you restrict to manufacturers, what about people who modify their vehicles? No amount of signing will be foolproof—tinkerers will find a way to extract signing keys. The SDV-to-SDV protocol could borrow from the contact tracing protocol used for COVID which requires about 60% of prior contacts for a positive match [§]. If SDV A says that at 11:00pm today it passed SDV B, C, D at location X, then if we were to reconstruct the blockchain from the perspective of each of SDV B, C or D, the collected data should be identical to the chain reported by SDV A, otherwise we would simply discard records that conflict with the longest chain, similar to how the blockchain mechanism works with Bitcoin. §: https://ncase.me/contact-tracing/ https://ncase.me/contact-tracing/

michaelt 6y ago

If a self-driving car needs data from other self-driving cars on the road, how's it going to work in the early stages, when only 0.1% of cars on the road have the feature? IMHO it'd be much easier to sell people a $10,000 self-driving option that worked immediately, than a $2,000 self-driving option and a promise it would start working at some point in the future, when xx% of vehicles on the road had deployed it.