Connect with us

Tech

AI’s memory crunch is coming for Android apps

Published

on

Google is making changes to its requirements for Android apps as developers face industry-wide memory chip shortages driven by the AI data center boom. This week, the tech giant announced two new app quality requirements, one of which is focused on reducing apps’ memory usage and optimizing its code.

The company explains that the mobile industry is now facing “significant hardware supply constraints that are altering device memory availability,” which can then, in turn, affect the consumer’s experience with their devices.

To address this, Google is now establishing new performance thresholds across several areas, like dynamic memory usage and bitmap usage. In addition, Google is adding code optimization requirements designed to prevent things like app slowdowns and crashes related to performance.

To aid developers, the company is rolling out new tools that will alert them when their apps exceed the new thresholds so they can take action by optimizing their code. More diagnostic tools will arrive later in the year, Google noted, including deeper insights provided by a Memory Limiter feature that prevents apps from using too much device memory.

The changes are meant to help Android app and game makers continue to deliver a quality experience, Google says, while reflecting a market where memory may not be as widely available, particularly for low-end devices where price points are a concern.

Developers have until February 2027 to meet the new thresholds, documented on Google’s Android Developer site.

In addition, the company will require all Play Store apps to meet the Zero Tap Sign-In requirement during device migrations by April 2027. This standard requires that apps with user sign-ins, whether optional or mandatory, automatically restore the user’s sign-in state when they move between Android devices using the Android Restore Credentials API.

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

>

Continue Reading

Tech

Hugging Face is selling a cute $399 open-source duck robot, Microduck

Published

on

Hugging Face unveiled the Microduck on Thursday, a cute little duck-like robot that sells for $399 — and ships before Christmas. 

Clem Delangue, CEO of Hugging Face, said the Microduck is an “open-source robot you can teach new tricks with reinforcement learning.” The 25-centimeter-tall duck can waddle, pick things up with its beak (up to 800 grams), get back up when it falls, crouch, and even roller skate.

“Welcome to the era of open-source affordable robots to democratize physical AI and world models!” Delangue said. 

Hugging Face is best known as a platform and community where developers go for open model weights. However, the company acquired French startup Pollen Robotics in April 2025 to build affordable, open-source AI hardware. A couple of months later, the two launched Reachy Mini, a small desktop robot. Today, they sell the $499 Reachy Mini, which is powered by a Raspberry Pi computer, and the $399 Reachy Mini Lite, powered by a Mac or PC. 

The new Microduck perceives the world with a camera, lidar sensors, and two IMUs (inertial measurement units that measure and report an object’s movement). 

Pollen Robotics noted that the Microduck’s behaviors can be trained in simulation and directly deployed on the robot. Developers can then fine-tune the bot, re-train it, and re-deploy it. The SDK (software development kit), simulation, and full RL training stack are available on GitHub.

For those who might find it concerning to have a robot with a camera in their bedrooms and private spaces, Delangue previously told TechCrunch that bots run by open-source models are much better from a privacy standpoint than “a black box system” controlled by a few organizations, “especially if these organizations’ CEO is not the most stable person in the world.” 

While open source gives developers auditability and control, it doesn’t provide a guarantee that sensitive data remains private after consumers start installing software applications on top of the model. Those apps can access the bot’s cameras and microphones and, depending on how they’re built, send that data to external services.

The launch of the Microduck comes as Hugging Face is reportedly set to be acquired by Nvidia at a $13 billion valuation. Nvidia and Hugging Face have been partners for years, with Nvidia providing the startup’s infrastructure since at least 2023. Both companies have also been publicly aligned in promoting open-source AI

Hugging Face was recently in the headlines over a cybersecurity incident that occurred after OpenAI’s systems breached its sandbox during safety testing and hacked into the platform’s servers.

TechCrunch has reached out to Hugging Face and Pollen Robotics for more information.

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

>

Continue Reading

Tech

This former PG&E engineer is building a ‘Google Maps for the underground’

Published

on

Josh Mackanic left a 10-year engineering career at Pacific Gas and Electric because he kept thinking about what the utility company doesn’t know about what’s buried underground.

“I had a job that got shut down because a pipe was in the excavation [that] we didn’t know about. Fortunately, we saw it before we impacted it,” he told TechCrunch. “But then there was this three-day ordeal of running around [asking] like: ‘Whose pipe is this? Can we tap it? Can we not? What’s in it?’”

Mackanic eventually got the answer, but it took three days, a delay that cost $60,000.

Despite being one of the largest utility companies in the United States, PG&E (and companies like it) only has visibility into the many miles of electric and gas lines it owns and operates. Pipes carrying sewage or water belong to other companies, and so does the information about where exactly they are. This leads to around 200,000 so-called “utility strikes” every year.

So in 2020, Mackanic founded a startup called CivilGrid designed to solve this problem. The company gathers disparate data regarding utility assets, property ownership, and environmental regulations and bundles it into what Mackanic cheekily refers to as “Google Maps for what’s underground.” It then sells access to governments, civil engineering firms, and utilities — including his former employer.

Now, Mackanic has raised a $26 million Series A to grow CivilGrid into something bigger. The round was led by Spark Capital, with additional investment coming from early-stage funds Afore, A*, Ford Street Ventures, and SNR. CivilGrid also received investment from Energy Impact Partners, a fund with a number of utility companies serving as LPs, which Mackanic pointed to as another vote of confidence from the industry.

“California’s energy needs are growing rapidly, and our customers expect us to deliver the infrastructure they need safely, reliably, and affordably,” Christine Cowsert, senior vice president of enterprise business and technology modernization at PG&E, said in a statement. “That means planning smarter from the start with tools like CivilGrid, which help our teams identify risks earlier, build more efficiently, and avoid unnecessary costs while keeping safety front and center.”

A case study performed by PG&E has already identified $60 million in avoidable paving costs across 1,600 planned gas distribution projects by using CivilGrid. Mackanic said he wants to bring on more customers and, eventually, start tackling other red tape problems that make it so hard to build things in the U.S.

“Right now, we’re providing engineers the data to be able to make a decision” about where to place infrastructure, he told TechCrunch in an exclusive interview. But “once I’ve given them the constraints, it’s not too hard for me to say, ‘Well, I would recommend that you put the pipe down right here, and oh, by the way, here are the permits you need in order to move to construction on this pipe. Would you like me to file those permits? Okay, let me file those permits.’”

Mackanic said there’s a “lot more” CivilGrid could start to automate “now that we have built this dataset and kind of captured the core user who’s at the very early stages of making decisions about what’s going to be built.”

Creating better visibility into the country’s subterranean infrastructure is not a novel idea, Mackanic told TechCrunch. But he said nobody had found a way to solve the problem of collecting, sorting, and securing the data, and striking up relationships with the partners most likely to pay for that information.

“One of the reasons I left PG&E to start CivilGrid was because I felt like this was a problem that was going to be better solved from the outside in than the inside out, and not because there’s [a] lack of awareness of the problem or even interest in solving it,” he said. “Utilities, they run relatively lean, right? Nobody wants to pay more for their gas bill than they have to.”

Like many kids, Mackanic said he spent much of his childhood dreaming of sending rockets into space. But even though he studied mechanical engineering in college and graduate school, he said he never expected to work for or with utility companies, digging into his own planet.

“I was like, aren’t these places boring?” he remembered thinking.

But working at PG&E made him realize just how high-stakes utility work can be — all so people can turn on their lights or start a pot of coffee every day without worrying that those things will work.

“The reality is, unfortunately, we do ourselves a disservice in not championing the extent to which they work wonders every day,” he said.

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

>

Continue Reading

Tech

Sprains, pain, and whiplash: Waymo and Zoox test drivers are getting hurt as robotaxis scale

Published

on

The rapid scaling of robotaxis over the last two years has come with a hidden human cost. 

Test drivers for Waymo and Zoox sustained more than two dozen injuries in 2024 and 2025 from hard braking or other sudden movements made by the autonomous vehicles, according to a TechCrunch review of data submitted to the Occupational Safety and Health Administration. 

These test drivers, who put autonomous vehicles through their paces on public roads, have suffered sprains, strains, whiplash, and more, according to OSHA data and interviews with former and current workers. Some have been sidelined for weeks or months.

Transdev, which employs and manages Waymo’s test drivers, reported 16 injuries tied to Waymo depot locations in three cities that were caused by the autonomous software braking hard or swerving suddenly. Zoox reported as many as eight worker injuries linked to hard braking. 

Other autonomous vehicle companies do not appear in the dataset, though it’s possible they’re exempt. The Zoox and Waymo injuries appear because the companies classify themselves as taxi, limo, or transit services, which OSHA subjects to greater disclosure because they’re “higher-hazard” industries. OSHA also only requires establishments with 100 or more employees in these industries to make this information public, meaning test drivers in newer cities where operations haven’t scaled may have been injured on the job without appearing in this dataset. 

TechCrunch was unable to learn whether Waymo or Transdev logged any new injuries in 2026. OSHA does not require companies to submit annual data until the following year as part of the agency’s Injury Tracking Application.  

In Zoox’s case, injuries are still happening, according to sources who spoke to TechCrunch on condition of anonymity.  

Two current and three former Zoox contractors told TechCrunch that the company’s test fleet vehicles — modified Toyota Highlander SUVs equipped with its self-driving system — were still braking hard and abruptly as recently as July. Three of those contractors said workers are still getting hurt. The workers were granted anonymity to discuss private operations. 

Zoox declined to answer specific questions about the injuries it reported last year or the claims contractors made to TechCrunch about the continued hard-braking issues. The company said it follows necessary reporting requirements and takes injuries seriously. 

Zoox also said hard-braking incidents are sometimes unavoidable and that the injuries reported last year represent a small percentage of the millions of miles its test fleet traveled. 

Waymo also declined to answer questions about the injuries reported to OSHA.  

“The safety of our riders, road users, and team is of paramount importance,” Waymo told TechCrunch. “As part of the process of responsibly deploying fully autonomous vehicles, we have driven tens of millions of miles with human operators behind the wheel. This is part of our ongoing validation process and learnings from these trips are integrated into the service.” 

Transdev declined to comment.  

More miles, more injuries 

WAYMO on SF freeway
Image Credits:Waymo

While Waymo has spent more than a decade developing its self-driving cars, it wasn’t until 2025 that it really started to scale. The company started offering rides in new cities and began mapping and testing in others, growing its commercial fleet to more than 1,500 vehicles by May 2025.  Today, that fleet is more than 3,500 vehicles.

Waymo doesn’t break out how many vehicles it uses just for testing. But the amount of autonomous vehicle testing almost certainly increased as the company expanded its service areas and started to familiarize itself with entirely new locations.  

Test driver injuries increased, too.  

Transdev reported five Waymo test driver injuries in 2024 across San Francisco, Los Angeles, and Phoenix. One was related to hard braking, while the other four were related to erratic vehicle behavior.  

In one case, a test driver reported that the car “suddenly backed up” while parking itself and “whipped the steering wheel,” catching his left hand and bending it “to the point he heard a crack in his wrist.” The employee spent more than two months away from work as a result. 

The number of injuries jumped to 11 in those three markets in 2025, and hard braking was the cause of nearly all of them. Last September, a test driver in Phoenix wound up having to spend 157 days away from work after their vehicle “made an exaggerated breaking [sic] event without any obstruction.” At least two other injury reports specifically mention that the hard braking happened for no apparent reason.  

The only injury report from 2025 that offers a reason for a hard-braking event involved an incident that happened in August in Los Angeles. The employee reported that “some kids were playing in the pathway.” The Waymo robotaxi stopped hard enough that the worker spent 175 days away from their job.  

“Software braked harshly” 

Image Credits:Zoox

The Zoox contractors said hard braking — or “brake jabs,” as they are referred to internally — can happen multiple times on a single test drive, sometimes in the same location. They can happen at any speed, though higher-velocity braking incidents typically take a harsher toll on the body, the workers said.  

The braking often happens when the autonomous driving system detects — or mistakenly detects — debris of some kind on the road, they said. 

The scariest incidents, the contractors said, are when a test vehicle experiences a “nogo” — internal code for an entire system shutdown that often leads to a brake jab.  

“You could be going 45 miles per hour and the car will enact a nogo, and suddenly you’re being whipped forward in the middle of the street, in traffic, and you need to be able to take over quickly,” one said. 

This behavior shows up repeatedly in the injury reports Zoox submitted to OSHA.  

On January 5, 2025, a Zoox test vehicle was driving autonomously in San Francisco when it abruptly slammed on the brakes. The “harsh” stop, as Zoox described it, injured the shoulder and upper arm of a contractor who was inside the SUV, leaving the worker on restricted duty for 16 days. 

Later that month, another driver was hurt after a Zoox test vehicle made an “unexpected” stop, injuring their neck and leaving them with a swollen shoulder. It took the contractor eight days to return to work without restrictions. 

Most of the Zoox-reported injuries appear to have happened to workers who sit in the driver’s seat of the SUVs. The company also reported an injury to an instructor who trains the test drivers. That injury took place on June 3 in San Francisco, when the test vehicle’s “software braked harshly,” causing the instructor in the back seat to report “sharp pain in lower right ribs and tightness in left shoulder upon returning to base.”  

The injured instructor reported additional injuries to the spine and pelvis, though their work restrictions lasted only four days, according to the data.  

Seven of the eight entries explicitly mention hard braking, nogos, or brake jabs. The eighth injury involved a worker who was sidelined for the longest amount of time — 46 days — and who reported “extreme back pain” after operating a test vehicle in Las Vegas in May 2025. Zoox wrote that the worker said they “were not sure if it was related to the behavior of the vehicle in autonomy or not but they said it hurts them to drive as well as take deep breaths,” and attributed the injury to a “vehicle maneuver.”  

Zoox is dealing with these hard-braking issues as it finally begins offering a true commercial robotaxi service in a bid to compete with Waymo. The Amazon-owned company started charging for rides in its purpose-built robotaxi for the first time this month in Las Vegas. It offers free rides in San Francisco and is testing in Atlanta, Austin, Dallas, Los Angeles, Phoenix, Seattle, and Washington, D.C. 

Four of the injuries Zoox reported in 2025 happened after the company issued a recall meant to resolve a year-long federal investigation into the tendency of its autonomous vehicles to brake unexpectedly. Zoox told TechCrunch the recall addressed an extremely narrow braking behavior and said the later incidents — and any resulting injuries — are unrelated to the software fix. 

Two of the people who spoke to TechCrunch said that braking can be challenging even when manually driving the Zoox test cars because of the extra weight of the self-driving sensor suite. Test drivers are trained to brake with the extra weight, they said, though they added that minor injuries can be sustained even during that training process. 

It’s hard to know if Zoox is experiencing hard-braking incidents with any regularity in its purpose-built vehicles, which do not have traditional controls such as a steering wheel or pedals. They either operate unoccupied or carry passengers who mostly aren’t employees – meaning any injuries wouldn’t be reported to OSHA.  

Zoox has reported two robotaxi crashes to the National Highway Traffic Safety Administration in the last eight months in which hard braking for debris resulted in the purpose-built vehicles being rear-ended. The company’s descriptions of the crashes don’t specify what debris was spotted in either case. 

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

>

Continue Reading

Trending

Copyright © 2017 Zox News Theme. Theme by MVP Themes, powered by WordPress.