Introduction

In recent years, autonomous driving has emerged as a meteoric rise, revolutionizing our understanding of transportation at an unprecedented pace.

However, its journey has not been without obstacles, marked by significant fluctuations and new transformations.

Unmanned Vehicle Coming (Official Account: Unmanned Vehicle Coming) engages with car enthusiasts to delve into this fascinating topic!

I. Past Years: Building Foundations and Ambitious Aspirations

Looking back, the nascent stages of intelligent driving resembled a meticulously crafted jigsaw puzzle.

High-precision maps, now ubiquitous, were non-existent. Network communication lagged behind today's standards, city launches were yet to commence, LiDAR popularization seemed distant, and the notion of "nationwide autonomous driving" was a mere dream, let alone end-to-end technology.

Practitioners in the intelligent driving industry, akin to engineers in other sectors, adhered to a top-down design blueprint, diligently contributing to this ambitious endeavor.

Each team harbored dreams, unwavering in their belief that they had a role to play in the vast landscape of intelligent driving, irrespective of team size or chosen technical approach.

II. Boom Years: Hype and Underlying Concerns

In 2020, the autonomous driving sector ignited like a barrel of gunpowder, with major players embarking on an aggressive expansion spree.

Excessive capital influx poured fuel on the fire, rapidly heating up the industry.

The talent market flourished, from Shanghai to Suzhou, Shenzhen to Guangzhou, and Hangzhou to Beijing, with a palpable thirst for intelligent driving professionals.

As autonomous driving transitioned towards mass production, it evolved beyond a mere technical issue into a vast, intricate system engineering challenge.

Perception, planning, testing, platforms, simulation, systems, and security emerged as key areas where teams focused their expansion efforts.

Teams acknowledged, "Post-mass production preparation, tasks have multiplied exponentially, necessitating substantial recruitment to enhance efficiency."

For a while, autonomous driving budgets soared, with headhunters fueling the frenzy.

Talented, average, and even PPT-proficient individuals boarded this seemingly promising gravy train.

Simultaneously, non-compete agreements proliferated, with restrictions once limited to high-level positions now encompassing ordinary engineers.

In some cases, engineers resorted to using pseudonyms for contracts and paying social security contributions through other companies to switch jobs.

However, beneath this apparent prosperity lurked a hidden crisis.

By 2024, the industry took a sharp downturn, and the once rampant expansion swiftly became a cumbersome burden.

III. Technological Shift: Opportunities and Challenges Amidst the Wave

The emergence of the Scaling Law in artificial intelligence, akin to a colossal stone tossed into the intelligent driving pond, stirred up substantial ripples.

In 2023, when Musk proposed that massive road test data could automate problem-solving and end-to-end autonomous driving technology emerged, many scoffed, deeming end-to-end autonomous driving unsafe and nonsensical.

Yet, just a year later, the tide turned dramatically. In 2024, end-to-end technology became the industry's recognized pinnacle, heralded as the future's hope. This swift transformation caught both insiders and outsiders off guard.

Amidst this technological shift, the highly efficient Ideal AD team stood out, reaping substantial benefits.

Intriguingly, the Ideal AD team was also one of the most cautious during the prior autonomous driving expansion spree, suggesting a subtle correlation.

With the rise of end-to-end technology, the demand for computing power in the autonomous driving industry surged.

XPeng asserted that the competition in autonomous driving lies in the cloud, while Xiaomi countered with a vast computing power reserve.

Conversely, RoboTaxi companies once at the industry's apex found themselves in a quandary.

On one hand, they struggled to keep pace with the end-to-end trend due to vehicle scale and computing power constraints;

On the other hand, industry trends compelled them to strive for catch-up.

Simultaneously, they hesitated over whether end-to-end technology could meet L4 mass production safety standards.

The once alluring vision of "nimble teams driving autonomous driving forward with sophisticated algorithms" appeared to dim under the end-to-end technology's influence.

To navigate this transformation, Pony.ai pioneered an attempt.

They believed that end-to-end imitation learning had limitations and could hardly surpass human capabilities. They advocated introducing a methodology akin to reinforcement learning to significantly evolve RoboTaxi capabilities.

Their approach involved building an interactive world model rather than relying on real-world sensor data, allowing algorithms to learn interactively in a virtual environment.

This mirrors NIOIN's proposed NSim.

While this path is fraught with challenges, it represents an inevitable choice for industry development.

From 2025, virtual simulation technology became the focal point of various teams' promotion efforts, with technologies like 3D reconstruction, world modeling, and Gaussian Splatting coming to the fore, unimaginable just a few years ago.

However, this new technology wave also caused industry turbulence.

Multiple autonomous driving teams initiated layoffs, whether RoboTaxi companies disappointed by capital or teams that proactively downsized after proving end-to-end technology.

Planning, control, testing, mapping, post-fusion, and other areas lost investment focus.

Leading competitors like Momenta, DJI Innovations, and Huawei began reassessing the significance of their internal R&D teams, as if the feast had ended, leaving only scraps.

IV. New Emerging Horizon: Dawn and Uncertainty of Embodied Intelligence

Fortunately, as the intelligent driving industry undergoes adjustment, a new horizon—embodied intelligence—quietly emerges.

This concept mirrors autonomous driving in 2016, with everything novel and uncharted.

By 2025, there is neither a consensus on methodologies nor toolchains within the industry. Each company harbors unique ideas, aiming to bring their hardware to market.

This means every solution has a chance to prevail, with valuable uncertainty intertwined with a promising future vision, attracting numerous investors.

Autonomous driving, with its broad user base, diverse application scenarios, and stable vehicle form, facilitates algorithm iteration.

However, embodied intelligence faces numerous uncertainties, such as the initial application scenario, whether the form must be humanoid, and how to cost-effectively collect data.

These issues deter impatient capital.

Despite the challenges, many intelligent driving legends have already dived in, their backgrounds encompassing nearly all current leading intelligent driving teams.

Many practitioners believe that years of autonomous driving experience can infuse certainty into this chaotic field.

In summary, Unmanned Vehicle Coming (Official Account: Unmanned Vehicle Coming) believes that intelligent driving development progresses sometimes slowly but steadily, and at other times rapidly and decisively. Judging by the 2025 trend, a brand-new industry feast appears imminent!

Let's observe and witness the autonomous driving field's new glory together.

What do you think, dear readers?