如何让交通信号灯更智能？

交通灯真是傻死了！

在等绿灯等得搓火的时候，你或许会一边用手指敲着方向盘，一边盯着一直不变的红灯这么暗自抱怨。

要让史蒂夫•史密斯博士说，这种看法字面上完全成立。

他解释说：“按常规交通管理系统的工作方式，你需要提前对定时系统进行预编程。但系统一安装好就会开始老化。而交通模式与流量一直在持续变化。”

随着人口与地形的变化，对一个十字路口进行重新编程的成本可能需要数千美元，因此，许多城市根本没有考虑这样做。那些令你陷入交通堵塞的信号模式，可能在三五年前便已过时。

由此产生的低效率会扩展到整体经济。德州农工大学交通研究所的《城市交通报告》发现，2011年，美国的交通拥堵成本总计为1210亿美元，而在1982年仅为240亿美元。

史密斯希望改变这种状况。他不仅是卡耐基梅隆大学的研究教授，也是初创公司Surtrac的老板。这家公司正在将先进的人工智能技术与规划应用于交通管理。Surtrac的智能交通信号依靠先进的传感器和强大的处理技术，每隔几秒钟（而非几年时间）就更新一次信号模式，以此响应交通流量的变化和意外事件，如交通事故或道路封闭等。自2012年以来，该公司一直在匹兹堡的东利伯蒂社区进行试点研究。

史密斯迅速指出，动态交通信号的理念并非他的发明。洛杉矶早在筹备1984年奥运会期间，便安装了最早的动态交通信号系统，现在该系统被称为“自动交通监控和控制系统”。目前，该系统仍在通过一台强大的中央计算机，协调超过4500个交通信号灯。这台计算机用越来越少的人力投入，做出及时的决策。

但史密斯认为，集中管理是洛杉矶系统的弱点之一。与之相反，Surtrac由每一个交通信号灯做出自己的交通管理决策。这些信号灯会与邻近的系统分享信息，但不存在集中控制，也没有主动的人为管理。

史密斯表示：“在试点研究开始几个月之后，穿过我们研究区域的一条主干道被切断。交通模式发生了巨大的变化——但我们没有做任何事情。”

结果显示，交通秩序变得高度协调。

史密斯解释说：“通过与邻近信号系统的通信，信息可以多跃程传播。因此，你可以通过一种分散式途径建立一个合理的长远规划。”

Surtrac信号合作的方式，与蚂蚁或鸟相同。这种方式被称为“虫群战略”，常被用于协调多架自动无人机执行各种任务，如调查研究和救援等。这种方式的优势包括可扩展性——额外增加信号灯非常容易，持久性——一个单元出现故障，剩余部分会自动调整以做出补救。

Surtrac在匹兹堡取得的成果非常显著。史密斯称，试点区域内的出行时间缩短了25%，这在很大程度上得益于空转时间减少了40%。而这也意味着排放减少21%。尽管Surtrac的结果仅来自24个交通信号灯，但它的效果远远超过洛杉矶系统减少交通拥堵16%的成果。试点得到了当地基金会和公私合营机构的资助，今年夏天将扩大规模，新增25个信号灯。

随着试点规模的扩大，Surtrac自适应技术的效果将得到最好的检验。匹兹堡首先将重新调整常规模型的信号灯计时，在Surtrac安装其智能系统的同时，该设置就处于运行状态。这意味着从一开始，这套智能系统就跟常规设置进行比较。

史密斯对结果充满信心。他说道：“传统观念认为，自适应信号不适合密集的城区。我们必将证明这种观念是错误的。”（财富中文网）

译者：刘进龙/汪皓

审校：任文科

Traffic lights are dumb.

Maybe you’ve muttered it to yourself, in a moment frustration, as you tap the steering wheel and stare at a persistent red.

As Dr. Steven Smith points out, it’s literally true.

“The way conventional systems work, you preprogram those timing systems in advance,” he explains. “But as soon as you install them, they start to age. Patterns and flows are continually evolving.”

Reprogramming a single intersection as population and geography shifts costs thousands of dollars—so some municipalities rarely get around to it. The signal patterns keeping you stuck in traffic may be three to five years out of date, if not more.

The resulting inefficiencies multiply into the broader economy. The Texas A&M Transportation Institute’s Urban Mobility Report found that in 2011, congestion cost the U.S. an aggregate $121 billion dollars, up from just $24 billion in 1982.

Smith is hoping to change that. On top of his position as a research professor at Carnegie Mellon, he’s the head of Surtrac, a startup that’s applying cutting-edge artificial intelligence and planning to traffic management. Surtrac’s smart traffic signals rely on advanced sensors and powerful processing to update their patterns, not every few years, but every few seconds, in response to shifting traffic volumes and unexpected events like accidents or road closures. They’ve been deployed in a growing pilot study in Pittsburgh’s East Liberty neighborhood since 2012.

Smith is quick to point out that he didn’t invent the concept of dynamic traffic signaling. Los Angeles installed one of the earliest such systems, now known as the Automated Traffic Surveillance and Control system, in preparation for hosting the 1984 Olympics. It’s still busy coordinating over 4,500 traffic lights through a powerful central computer, which makes timing decisions with less and less human input.

Smith is convinced, though, that centralization is a weakness of systems like the one in Los Angeles. Instead, Surtrac lets each signal make its own traffic management decisions. Though the signals share information with their neighbors, there’s no central control, and no active human management.

“A few months after we originally installed the pilot, one of the major arteries through our area got cut off,” Smith says. “Traffic patterns changed completely—we didn’t do a thing.”

What emerges, though, can look highly coordinated.

“By communicating with your neighbors, information can propagate multiple hops,” explains Smith, “So you can actually end up building a reasonably long-horizon plan in a distributed way.”

Surtrac’s signals cooperate in much the same way groups of ants or birds do in nature. This approach, sometimes referred to as “swarming,” has become frequently discussed as a way to coordinate groups of autonomous drones for tasks such as surveying or search and rescue. Advantages of the approach include scalability—additional lights can be added very easily—and durability—one unit can go down, and the rest will automatically adjust to compensate.

The results Surtrac has achieved in Pittsburgh are dramatic. Smith says travel times in the pilot area have been reduced over 25%, thanks mostly to a 40% reduction in idling. That also translates into a 21% emissions reduction. Those results significantly edge out the 16% congestion reduction achieved by the Los Angeles system, though Surtrac’s results are from only 24 traffic lights. The pilot, funded by local foundations and public-private partnerships, will be expanded this summer to include 25 more signals.

With the expansion, Surtrac’s adaptive technology will get the best test so far of its effectiveness. The city will first be re-timing the lights on a conventional model, and that setup will run while Surtrac installs its smart system. That means they’ll be measured against a conventional setupat its freshest.

Smith is confident of the outcome. “A lot of the conventional wisdom is that adaptive signals can’t work in dense urban areas,” he says. “We’re definitely proving that wrong.”