LNG: TESLARATI

Roundabouts aren’t always the easiest to navigate for human drivers, but one Tesla Model 3 was able to successfully go through one using its Smart Summon feature.

The host of YouTube channel All Electric wanted to see how his Model 3 would perform at navigating a roundabout using only Smart Summon. His Model 3, which had Tesla’s 2019.40.50.7 update installed, was able to make its way around the circular junction with minimum hiccups.

As the All Electric host pointed out, his Model 3 was able to enter the roundabout smoothly, while going at 5 mph and slowing down in some parts of the maneuver, before carefully exiting to reach its intended target location.

Watch how this #Tesla Model 3 safely takes a roundabout autonomously (via smart summon, not FSD…yet) : pic.twitter.com/DZCcrBqR4t

— TESLA_saves_lives (@SavedTesla) January 29, 2020

Tesla’s Smart Summon is a feature that allows users to remotely operate their Tesla from a parking spot to a designated location using their vehicles’ Full Self-Driving suite. The feature still requires users to be within 200 feet of their car, and uses the vehicles’ system of cameras, radar, and ultrasonic sensors to get around obstacles. Smart Summon is a step closer to Tesla CEO Elon Musk’s vision of deploying a fleet of autonomous vehicles.

Smart Summon has received plenty of improvements since it was released in September last year as part of the V10 update. The feature has also been put to the test many times, most recently proving that it can drive through a torrential downpour to rescue a family, navigate an intense obstacle course, and take its owner through a drive-thru restaurant.

With full backing from regulatory bodies, Tesla will likely be able to remove the 200-feet restriction on Smart Summon and allow users to call their cars from greater distances. CEO Elon Musk once even mentioned using Summon to call vehicles from across the country. Ultimately, Smart Summon may be a small feature for now, but it will play a central role in the deployment of self-driving Robotaxis.

Musk revealed further details for his vision of an autonomous ride-sharing program during Tesla’s Autonomy Day last year. Musk said he hoped to get the Tesla Network’s Robotaxi service running by 2020.

Watch All Electric‘s latest Smart Summon test in the video below.

(adsbygoogle = window.adsbygoogle || []).push({});

Elon Musk confirmed a future update will allow Tesla vehicles to create micro maps of roads that will help owners avoid potholes or other changes in the environment that are not available in conventional maps. The micro maps created from data collected by its fleet will be shared with fellow Tesla owners when they drive along the same road.

Reto Siegrist from Switzerland asked the Tesla CEO on Twitter about the possibility of adding the micro map feature to which Musk replied with a simple, “Yes.” Potholes can be big headaches for Tesla owners because they can cause serious accidents, damage tires and rims, which need repairs that can instantly burn through a few thousand dollars.

Yes

— Elon Musk (@elonmusk) February 3, 2020

According to TRIP, a national transportation research non-profit group, about a third of major urban roads in the United States have pavements that are in substandard condition. Driving on such deteriorated roads cost motorists around $130 billion.

“Drivers are paying a hefty price for our nation’s crumbling roads and bridges. Those traveling daily through urban cities bear the weight of the problem – with many wasting thousands of dollars each year on rising transportation costs due to pot holes and wasted fuel,” said Kathleen Bower, AAA senior vice president of public affairs and international relations

Elon Musk has previously mentioned an anti-pothole feature that will allow Autopilot to safely avoid potholes without the need for driver input. This micro map shared among Tesla vehicles on the road will be a welcome addition to the anti-pothole feature.

Definitely

— Elon Musk (@elonmusk) April 7, 2019

Detecting potholes can actually be challenging because unlike other obstacles, they are depressions on the road and may present differently depending on the current weather and lighting condition. In an ideal situation, the camera of Tesla’s electric car can determine a pothole by comparing the surface irregularity with the rest of the road and the texture inside the pothole compared to nearby pavement.

Tesla shared how its team is making progress in terms of developing its Full Self-Driving capabilities.

“…in terms of labeling, labeling with video in all eight cameras simultaneously. This is a really, I mean in terms of labeling efficiency, arguably like a three order of magnitude improvement in labeling efficiency,” Musk said.

The detection of potholes and the ability to seamlessly negotiate such obstacles on the road will be one big step for Tesla’s dream of having a fleet of truly autonomous cars. The ability of each car to share this information with other Teslas is also making most of the technology to make roads safer for everyone.

“Our networks learn from the most complicated and diverse scenarios in the world, iteratively sourced from our fleet of nearly 1M vehicles in real-time. A full build of Autopilot neural networks involves 48 networks that take 70,000 GPU hours to train. Together, they output 1,000 distinct tensors (predictions) at each timestep,” Tesla wrote to describe how its Autopilot neural networks work.

Elon Musk is also making sure that Tesla can accelerate further development of its FSD and Autopilot and will host an AI party at his house in a few weeks’ time. During the said event, Musk is hoping to find new talents to join Tesla and make functional applications out of the data available to the company.

(adsbygoogle = window.adsbygoogle || []).push({});

Thanks to a recent cluster of major milestones, SpaceX’s family of Falcon 9 and Heavy rockets are rapidly nearing the halfway point along the path to several ambitious goals for booster and fairing reusability.

Back in the early 2010s, SpaceX’s CEO Elon Musk’s original dream was to make Falcon 9 and Falcon Heavy 100% reusable, meaning that the company would need to find ways to reliably recover boosters (first stages), payload fairings (or Dragon spacecraft), and the rocket’s upper (second) stages. The concept of Falcon 9 second stage reuse actually survived all the way into 2018 before Musk ultimately conceded defeat, accepting that Falcon 9 and Heavy simply didn’t offer the performance necessary to make full reusability a worthwhile investment. The concept, however, still lives on in SpaceX’s next-generation Starship launch vehicle.

This does mean Falcon rockets will never be fully reusable, but it’s still up to SpaceX to decide how far they’ll push the envelope with the rockets’ existing reusable hardware. At the moment, it appears that a vast majority of Falcon rockets will be able to be routinely recovered and reused, capitalizing on the fact that Falcon 9 and Falcon Heavy boosters already represent some 50-75% of the cost of building each two-stage rocket. While Falcon upper stages and Dragon trunks will never be reused, both booster and payload fairing reuse are rapidly approaching their own unique halfway points on the path to ambitious reusability targets.

Shortly after SpaceX’s January 29th Starlink V1 L3 launch, carrying the third batch of 60 upgraded v1.0 satellites to orbit, twin fairing recovery ships GO Ms. Tree (formerly Mr. Steven) and Ms. Chief teamed up for their second-ever simultaneous fairing catch attempt. Ms. Chief – only active since November 2019 – reportedly just barely missed her first successful catch, while Ms. Tree managed to snag one of the Falcon 9 fairing halves in her massive net – the ship’s third successful catch.

Worth an estimated $3M per half according to CEO Elon Musk, Falcon 9’s payload fairing represents approximately 10% of the rocket’s total manufacturing cost. Made out of a carbon fiber and aluminum honeycomb composite material, fairings also also takes a disproportionate amount of time and space to produce – primarily due to their large size (a school bus could comfortably fit inside a fairing) and the need for commensurately large curing ovens. That composite honeycomb structure also makes it relatively easy for Falcon payload fairings to suffer from corrosion when dunked in seawater, leading SpaceX to the seemingly bizarre solution of installing giant arms and nets on ships.

Catching fairings has proven to be incredibly unforgiving, however, and SpaceX has simultaneously worked to make its Falcon fairings much more waterproof (and thus resistant to corrosion) while keeping them as light as possible. In fact, SpaceX’s first fairing reuse occurred less than three months ago and used two halves that previously landed in the Atlantic Ocean, demonstrating that difficulties reliably catching fairings will not stand in the way of reuse.

Ms. Chief missed her January 29th catch attempt, she still managed to fish her fairing half out of the ocean, while Ms. Tree’s successfully-caught half means that SpaceX ultimately recovered the full Starlink V1 L3 fairing. With a little luck, that recovered fairing will launch again in the near future.

Five for 5

Simultaneously, SpaceX is making excellent progress along the path to airliner-like rocket reusability. In November 2019, on the same Starlink mission that debuted flight-proven fairings, Falcon 9 booster B1048 became the first SpaceX rocket to launch (and land) four times. Less than two months later, Falcon 9 B1049 doubled down on that reusability milestone, becoming the second booster to launch and land four times, followed by Falcon 9 B1046 just 12 days later. Falcon 9 B1046 was (intentionally) destroyed after its fourth launch, precluding a fourth landing attempt, but it emphasizes just how confident SpaceX is in Falcon 9’s Block 5 upgrade.

Designed to allow each Falcon 9 and Heavy booster to perform a minimum of 10 launches and landings, the Block 5 upgrade is potentially just a few weeks away from reaching the halfway point along the path to that ambitious reusability design goal. Speaking at the NASA Kennedy Space Center earlier this month, a SpaceX engineer recently revealed that a Falcon 9 booster would conduct its fifth launch in support of a Starlink mission (either Starlink V1 L4 or L5) scheduled no earlier than (NET) mid-to-late February.

Pictured above, Falcon 9 booster B1048 – the first to launch four times – is the likeliest candidate for the first fifth flight of a SpaceX rocket. If the booster’s reuse goes as planned, it’s safe to say that Falcon 9 B1049.4 will follow closely on the heels of its predecessor with its own fifth-flight milestone. All things considered, SpaceX’s workhorse rocket is rapidly approaching the zenith of its theoretically-achievable reusability.

Check out Teslarati’s newsletters for prompt updates, on-the-ground perspectives, and unique glimpses of SpaceX’s rocket launch and recovery processes.

(adsbygoogle = window.adsbygoogle || []).push({});

SpaceX has completed its third rocket launch of 2020 and the most recent booster to launch safely returned to Port Canaveral on Saturday after an exceptionally hard drone ship landing.

Falcon 9 booster (first stage) B1051 lifted off for the third time on January 29th, following up two prior orbital-class missions by placing SpaceX’s fourth batch of 60 Starlink satellites into low Earth orbit (LEO). B1051 debuted on March 2nd, 2019 when it became the first Falcon 9 rocket to launch SpaceX’s next-generation Crew Dragon spacecraft, successfully sending the vehicle on its way to what would end up being a flawless rendezvous with the International Space Station (ISS). Less than four months later, B1051 completed its second mission, this time lifting off from SpaceX’s Vandenberg Air Force Base (VAFB), California facilities before landing in zero-visibility fog conditions just a thousand feet from the pad.

Compared to some of the higher-energy geostationary (high orbit) launches SpaceX often performs, B1051’s two prior launches allowed for relatively gentle reentries and landings. On January 29th, 2020, after sending SpaceX’s 3rd batch of upgraded Starlink v1.0 satellites (Starlink V1 L3) on their way to space, the Falcon 9 booster experienced the hardest successful landing seen after a SpaceX launch in quite some time.

With Starlink V1 L3 complete, SpaceX has officially launched an incredible 120 satellites weighing some 32 metric tons (70,500 lb) in a single month – 22 days, to be precise. If everything goes as planned, those two monthly Starlink launches should become SpaceX’s average over the rest of 2020, necessary to satisfy the company’s goal of completing 20-24 Starlink launches this year alone. If SpaceX replicates its January successes this month, the company’s Starlink constellation – already ~230 satellites strong – may even be ready to start serving internet to customers in the northern US and Canada as early as March 2020, less than two months from now.

Meanwhile, the mission marked SpaceX’s second Falcon 9 landing and recovery of the new year, as well as the sixth time an orbital-class SpaceX booster has completed three launches. SpaceX continues to push the envelope of reusable rocketry ever since it debuted Falcon 9’s Block 5 upgrade in May 2018.

Designed to enable no less than 10 launches per booster with minimal refurbishment in between, SpaceX’s Block 5 reusability milestones have gotten much closer together ever since the company began dedicated Starlink launches, reusing a payload fairing for the first time and launching two Falcon 9 boosters for the fourth time in just the last two and a half months. In fact, SpaceX already has plans to launch Falcon 9 booster B1048 for the fifth time – another major reusability first – as early as the next 4-5 weeks.

Hard landing; tough rocket

Starlink V1 L3’s launch followed a trajectory almost exactly identical to the two V1 missions that preceded it in November 2019 and January 2020 and Falcon 9 B1051 ignited its central Merlin 1D engine for the last time around eight minutes after liftoff. Twenty seconds or so later, the Falcon 9 booster rapidly shut down its landing engine, visibly falling several feet onto the deck of drone ship Of Course I Still Love You (OCISLY).

The results of that unintentionally hard landing are extremely apparent in photos taken of the same booster after its first (March 2019) and third (Jan 2020) landings on drone ship OCISLY, compared above. Taken from almost identical perspectives as the drone ship passed through the mouth of Port Canaveral, the difference in the booster’s height and stance are hard to miss, with B1051’s engine bells and the black ‘belt’ of its heat-shielded engine section clearly sitting several feet lower after Starlink V1 L3.

While subtle, the most important difference is near the tips of each visible landing leg’s telescoping boom, visible in the form of a final, smaller cylinder on the left (earlier) image. On the right, that cylinder has effectively disappeared. This is actually an intentional feature of Falcon 9’s landing leg design: known as a ‘crush core’, the tip of each leg boom holds a roughly 1m (3ft) long cylinder of aluminum honeycomb, optimized to lose structural integrity (crush) only after a specific amount of force is applied. In essence, those crush cores serve as dead-simple, single-use shock absorbers that can be reused as long as a given booster’s landing is gentle enough.

B1051’s third landing was definitely not gentle enough, but it appears that the booster’s rough fall onto the drone ship’s deck was just within the safety margins those crush cores provide. Why B1051 fell onto the deck is unclear, potentially caused by the drone being at the bottom of a swell or a last-second anomaly with the booster’s landing engine. Thankfully, regardless of the cause of the anomaly, B1051’s crush cores can be quite easily replaced, meaning that the booster can remain operational as long as its hard landing didn’t cause any less-visible damage or stress elsewhere on the rocket.

In short, SpaceX smart design decisions very likely allowed a part worth just a few thousand dollars to save a Falcon 9 booster worth tens of millions of dollars from the scrap heap. With a little luck, B1051 should have at least several more launches in its future before entering retirement.

Check out Teslarati’s newsletters for prompt updates, on-the-ground perspectives, and unique glimpses of SpaceX’s rocket launch and recovery processes.

(adsbygoogle = window.adsbygoogle || []).push({});

Elon Musk announced that Tesla will be hosting an AI hackathon, together with the company’s artificial intelligence and autopilot team, at his house in four weeks’ time.

The Tesla chief announced his plans via Twitter on Sunday. Despite impressive numbers revealed during the Q4 2019 earnings call and update, Musk and his Tesla team are not resting on their laurels and remain focused on pursuing advancements to its neural network, which is in the center of Tesla’s goal of achieving a full self-driving vehicle.

During the recent Q4 earnings call, an investor asked the Tesla chief executive for updates on FSD.

“I think that’s looking like maybe it’s going to be couple of months from now. And what isn’t obvious regarding Autopilot and Full Self-Driving is just how much work has been going into improving the foundational elements of autonomy,” Musk said.

Tesla will hold a super fun AI party/hackathon at my house with the Tesla AI/autopilot team in about four weeks. Invitations going out soon.

— Elon Musk (@elonmusk) February 2, 2020

Musk continued to explain how the Tesla team is making great strides in labeling efficiency.

“…in terms of labeling, labeling with video in all eight cameras simultaneously. This is a really, I mean in terms of labeling efficiency, arguably like a three order of magnitude improvement in labeling efficiency. For those who know about this, it’s extremely fundamental, so that’s really great progress on that,” Musk said.

Tesla vehicles rely on a custom chip that boasts of 144 tera operations per second (TOPS) for its self-driving capabilities. This two-chip FSD computer works in tandem with LPDDR4 RAM modules that come with a peak bandwidth of 68 GB/s. There are also two neural network accelerators that work in tandem to process as much as 1TB of data per second. This setup is roughly three times faster, about 80%, and about 1.25 times more power-efficient than the previous hardware. It is also able to process about 2,300 frames per second compared to the 110 frames per second processed by Tesla’s Hardware 2.5.

In his series of tweets on Sunday, Musk also mentioned Tesla’s “Dojo” supercomputer, which is speculated to be capable of processing vast amounts of data to train the company’s neural network. Through active learning, Tesla curates the most useful video clips from its fleet of connected cars and train the neural net to recognize things that it did not previously know.

“Our networks learn from the most complicated and diverse scenarios in the world, iteratively sourced from our fleet of nearly 1M vehicles in real-time. A full build of Autopilot neural networks involves 48 networks that take 70,000 GPU hours to train. Together, they output 1,000 distinct tensors (predictions) at each timestep,” Tesla wrote on the Autopilot AI section of its website.

“At Tesla, using AI to solve self-driving isn’t just icing on the cake, it the cake” – @lexfridman

Join AI at Tesla! It reports directly to me & we meet/email/text almost every day. My actions, not just words, show how critically I view (benign) AI.https://t.co/iF97zvYZRz

— Elon Musk (@elonmusk) February 2, 2020

The last major software update rolled out by Tesla allowed its vehicles to visualize more things while driving in inner-city streets. Teslas now render stoplights, stop signs, traffic cones, traffic pylons, and more.

With the upcoming AI hackathon, Tesla will get together with developers to seek out more efficient algorithms and overall improvements to the core logic for its Full Self-Driving suite through a time-boxed event. With fresh eyes working with the existing AI and autopilot team of Tesla, the carmaker may be able to accelerate the timeline and rollout of its full-featured Full Self-Driving suite sooner.

Further advances in FSD and its Autopilot feature will widen the gap between Tesla and its competitors and solidify the company’s position as one of the leading automakers in the world. These improvements will also take Tesla a step closer to the possibility of Robotaxis that they can deploy at scale.

The hackathon will also allow Tesla to fish for new AI talents to join the team. On Sunday, Musk also mentioned that the electric carmaker is looking for world-class chip designers and C++/C engineers for vehicle control and other functions of Tesla vehicles.

Musk reiterated that educational attainment is not important when joining Tesla but rather a clear understanding of how AI and neural networks function and the ability to build useful applications using that knowledge.

(adsbygoogle = window.adsbygoogle || []).push({});