FSD's Emergency Avoidance — Sometimes a Ghost, Sometimes a God

Yesterday I watched a real-time dashcam video of a Tesla making an emergency swerve to avoid a car that suddenly shot up from the left lane entrance ramp. My immediate thought: human reaction speed simply can't handle that.

In that situation, most of us instinctively slam the brakes — which on a highway is itself dangerous. Being able to safely dodge to the right lane like FSD did is clearly the better strategy. Unfortunately, most human drivers just can't pull it off.

After driving with FSD for a long time, you develop a very strange kind of trust.

Not that it's always right. Not that you always understand why it did what it did.

But you realize: many of those heart-stopping emergency maneuvers that made you break out in a cold sweat — when you replay them later, most of them genuinely protected your safety.

Over all my years of manual driving, my default in emergencies was always the reflexive hard brake. Because only by slowing down did I feel any sense of control. It wasn't that I didn't know how to steer — I was afraid to. Because you have to check: is the right lane clear? Is there a car in my blind spot? How fast is the car behind me? Is the other driver a novice? Are they panicking? This entire judgment chain is serial — the human brain simply can't process it fast enough.

So most people, like me, instinctively hit the brakes.

But FSD is different. It's not just that it has watched countless expert drivers — it's more like a driver with many sets of eyes and reaction speeds many times faster than ours. It's constantly watching all four directions, constantly computing the space, speed, and risk in every lane.

That's why sometimes, it dares to execute lane escapes that we wouldn't dare attempt.

Of course, this brings another problem: sometimes it's overly cautious. A bird suddenly flies past in front — it might trigger an avoidance reaction. And some emergency dodges, even in hindsight, we may not fully understand. The infamous "phantom braking" from a few years ago is the classic example: tree shadows, bridge shadows, lighting changes, even road texture could trigger false alarms.

But here's what's remarkable: phantom braking has almost disappeared in recent years. I've barely encountered it myself in over a year. This tells us it's no longer just "seeing something that looks like danger" — it's increasingly understanding: what will actually hit me, and what is merely a shadow.

This is the most fascinating thing about FSD.

In its early days, it sometimes acted like a clumsy student. Now it behaves like an inhumanly fast-reacting entity.

Yesterday it executed one particular avoidance maneuver that I didn't fully understand either. Maybe it overreacted. Maybe it saw a risk we didn't. But I'm not going to dig deeper into it.

Because after long-term use, my trust in it doesn't come from faith — it comes from replaying every drive, time after time.

The vast majority of the time, those tense maneuvers that felt excessive in the moment — looking back, they were protecting us. It is far more cautious and safe than this old-timer-among-clumsy-drivers.

And that's enough.

What will truly transform driving in the future may not be whether it can drive like a human.

It's that it finally can drive unlike a human.

发布者

立委

立委博士,多模态大模型应用咨询师。出门问问大模型团队前工程副总裁,聚焦大模型及其AIGC应用。Netbase前首席科学家10年,期间指挥研发了18种语言的理解和应用系统,鲁棒、线速,scale up to 社会媒体大数据,语义落地到舆情挖掘产品,成为美国NLP工业落地的领跑者。Cymfony前研发副总八年,曾荣获第一届问答系统第一名(TREC-8 QA Track),并赢得17个小企业创新研究的信息抽取项目(PI for 17 SBIRs)。

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注

这个站点使用 Akismet 来减少垃圾评论。了解你的评论数据如何被处理