Tesla offered a free trial of its FSD system in December. I was driven autonomously for most of the month and came away with new insights.

The latest software was impressive, but my experience revealed a glaring problem that could hamper Tesla’s path toward a full, unsupervised Robotaxi service.

First, the good stuff. I have a Model 3 Performance with HW4 hardware. It doesn’t have a camera near the front bumper, but otherwise it’s modern hardware. I also have an up-to-date version of FSD.

A new state of driving

This combination resulted in an amazing FSD experience. I drove from Silicon Valley to Lake Tahoe in supervised autonomous mode and had no issues. I drove into SF and back several times, trouble-free.

I estimate 90% to 95% of my driving was in FSD mode during December. It’s a new state of driving that I think is safer than human driving. I spend less time worrying about where I’m going and which turn to make. Instead, my eyes are on the road and my hands are ready near the wheel, just in case.

Despite the ability to text while being driven now, I pay more attention to the road in FSD mode, not less.

A surprise bonus: I argue (OK, debate!) less with my wife about directions when we are on FSD-powered trips together. There’s no discussion about whether this or that route is better. FSD just does it. This reduces the potential for marital spats. That’s probably worth the entire $100-a-month FSD subscription cost.

When Tesla told me my month-long trial was ending, I was genuinely disappointed. Do I pay monthly now? No. But I don’t drive enough at the moment. If my wife’s or my commute becomes much longer, we would certainly pay.

That glaring problem

This was all supervised autonomous driving. That means Tesla requires human owners to be at the wheel and alert at all times.

Unsupervised autonomy is the next frontier. That’s when cars drive themselves with no human at the wheel. Waymo already does this in Silicon Valley and a few other urban areas in the US. Tesla’s Robotaxi service started doing this in Austin recently, but elsewhere, these Robotaxis still have safety drivers behind the wheel.

Making this leap is a huge feat. And Tesla is trying to do this with a purely camera-based system, while Waymo uses cameras, Lidar, and other sensors that are not just vision-based.

This is the big schism in the industry. Backers of the multi-sensor approach say their systems are better and safer because components such as Lidar collect data without “seeing,” so they work in more environments. However, this is more expensive than camera-only systems.

Tesla’s approach is cheaper, so that’s good for efficiently mass-producing millions of robotaxis, such as the upcoming Cybercab. But, can camera-only systems actually work as well in unsupervised situations?

This is where my December FSD experience sowed some seeds of doubt in my mind. To be clear, I’m not an expert. I’m a huge fan of my Tesla vehicle, and I hope both autonomous approaches work and make roads safer. Still, here’s what happened last month.

The main front camera at the top of my windshield sometimes gets condensation, dust, or dirt inside the glass and plastic compartment.

A couple of times, when I was in FSD mode and being driven into the low sun, I got a loud, red warning sign on the screen of my Tesla telling me to take over immediately or the system would pull over. I took the wheel quickly with no further issues and restarted FSD soon after.

But these incidents were slightly unsettling. My interpretation is that sunlight was streaming into the front camera compartment, and the glare was combining with condensation and/or dust to degrade the quality of the images being collected.

A week or so later, I got an alert from Tesla to bring my vehicle in for a service. The goal was to clean the inside of the front camera. I set up an appointment, which the Tesla app told me would cost $60.

I took it to a nearby Tesla location, and the cleaning lasted about an hour. Tesla didn’t charge me. After that, I didn’t get any red warning signs for the rest of December while using FSD.

This front camera is not easy to clean because it has a housing around it, which might also trap moisture that could otherwise dissipate more easily from the car’s interior.

If Tesla wants its cars to be fully unsupervised Robotaxis, a front interior camera that fogs up and needs to be professionally cleaned seems like a potentially big problem. With a million Robotaxis on the road, how many times might Tesla service employees have to clean these cameras?

This also suggests that a multi-sensor approach — that doesn’t 100% rely on cameras that need cleaning — might be a better way to attain full autonomy at scale.

“A major area of focus”

There are signs that Tesla may already be responding to this issue, as it gets more experience running its Robotaxi service.

In September, Elon Musk wrote on X that keeping the front camera clean is “a major area of focus.”

About a month later, the company rolled out a software update that enabled the windshield wipers to do extra wiping around the front camera. You can see it in action here. That only addressed exterior dirt and moisture, though.

More recently, Tesla Robotaxis have been spotted with new cleaning systems that squirt water onto the outside cameras to clean them.

More importantly, there’s a new setup for the front camera behind the windshield, according to photos posted on X recently by Tesla investor and owner Sawyer Merritt.

“Black lines surrounding the entire front camera housing, which could mean improved sealing to prevent off-gassing inside the car from fogging the cameras, potentially eliminating the need for Tesla to perform front camera cleanings (just speculation),” he wrote.

Another Tesla owner posted a photo this week of a new Model Y with an updated front camera housing.

Tesla also has a patent for a new way of preventing front camera glare, for use in autonomous and semi-autonomous vehicles.

“The glare shield features a textured surface composed of an array of micro-cones, or cone-shaped formations, which serve to scatter incident light in various directions, thereby reducing glare and improving camera vision,” Tesla wrote in the patent filing.

It’s unclear how well these fixes will work. But Tesla is obviously trying hard to address these challenges. Its camera-only approach depends on it.

I asked Tesla and Elon Musk all about this on Monday in a detailed email. They didn’t respond by publication time on Tuesday morning.

Sign up for BI’s Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.