3 ms·
Not just Tesla that gets confused. Other cars with lane assist also get easily confused if there’s markings on the road. Pretty sure my Passat would do the same
by dawnerd 8y ago
Not just Tesla that gets confused. Other cars with lane assist also get easily confused if there’s markings on the road. Pretty sure my Passat would do the same thing.
- cheeze 8y agoAgree, but Tesla is the poster-child for "self-driving" which certainly makes something like this that much more alarming.
- ccnafr 8y agoI thought Waymo (Google cars) was the poster-child for self-driving cars
- darkpuma 8y agoOne has better tech, the other has bolder marketing.
- imtringued 8y agoTesla's autopilot is very well known to cause deadly accidents.
- ironmagma 8y agoIs it? I would say Waymo is a lot closer to being the poster child; a significant number of people have realized that Tesla isn’t true self-driving despite their claims. Disclosure: I work at Cruise (a self-driving car company)
- cma 8y agoTesla's is supposed to be more than lane assist, something like only regulators keeping it from being released by now. When combined with the big neural link announcement that was supposed to happen weeks ago, you may be able to monitor autopilot while you sleep by having it tap into your dreams. They are also going to give us point-to-point earth rocket travel for less than a first class ticket within 8 years. I'm joking about neural link, though it did miss an announcement, but I'm not joking about the rocket one.
- kerng 8y agoSame for humans too! Not trying to state the obvious but the Fig 31 and 32 in the paper might also make humans believe that there is, or should be a lane - although it looks like an error. The changes on the road are certainly not minor at all! What is critical is to consider further context and its unclear if the interpretation of the image by itself would cause the AV to make an incorrect decision. It's part of a large amount of information that is consumed in the decision-making process I assume - again similar to humans.
- mannykannot 8y ago> its unclear if the interpretation of the image by itself would cause the AV to make an incorrect decision. That's exactly what was achieved.
- mannykannot 8y agoTesla is particularly vulnerable to this example because of its dependence on vision systems.
- PinguTS 8y agoLane markings is always about vision system. The brand does not matter for this.
- hellllllllooo 8y agoHow you implement these systems matters. GM super-cruise (not to be confused with Cruise automation) uses HD mapping + GNSS + visual queues and could potentially do better due to having priors from the map. Tesla's approach that doesn't utilize a map is likely more susceptible. https://www.wired.com/story/cadillac-super-cruise-ct6-lidar-laser-maps/ https://www.wired.com/story/cadillac-super-cruise-ct6-lidar-... https://www.cadillac.com/world-of-cadillac/innovation/super-cruise https://www.cadillac.com/world-of-cadillac/innovation/super-...
- LeoPanthera 8y agoGM super-cruise is the only system to work like this. (Available in exactly one model of car.) Meanwhile there's ~10 other car manufacturers out there making lane-keeping systems that use cameras, and yet people only care about bashing on Tesla.
- hellllllllooo 8y agoGM super-cruise is the best system out there right now and they are very modest in their claims. All these systems have problems. Tesla is the one making the big claims and drawing attention to themselves. How many others are claiming full autonomy is just around the corner? They bring the publicly on themselves and can't have it both ways. https://electrek.co/2019/03/29/tesla-full-self-driving-computer-retrofit-elon-musk/ https://electrek.co/2019/03/29/tesla-full-self-driving-compu...