在COMVEC 17 SAE 商用车大会的开场主旨演讲中，博世集团商用车和工程机械部门总裁Johannes-Joerg Rueger博士表示，完全自动驾驶系统大约需要1亿行代码，差不多是波音787梦想客机的6倍。对此，Rueger还调侃道，“看起来，脚踏实地难于上青天啊。”其实他的观点很明确——要打造出全自动驾驶商用车，是一项极其艰巨的任务。

对此，Rueger同与会来宾进行了交流，就自动驾驶的趋势、益处以及解决方案等问题分享了他的看法，尤其谈到了用于公路驾驶的自动驾驶辅助系统（ADAS）以及公路与非公路车辆所采用的自动驾驶技术。

SAE：商用车是否比乘用车更需要自动驾驶技术？

Rueger：商用车有可能比乘用车更需要采用自动驾驶技术。德国的相关数据显示，在由重型卡车引起的有伤亡交通事故中，如果借助已有的科技手段——包括紧急制动以及变道预警等，有一半以上是完全可以避免的。而另有40%的事故，也有望通过几年后上市的驾驶辅助系统化险为夷。而除了安全方面的需求之外，没有事故也意味着道路将更加通畅。而对于乘用车来说，自动驾驶可以带来更多的方便。如果在驾驶过程中不用一直紧握方向盘，而是可以腾出手来做些其他的事，那就太棒了。

但从经济效益的角度来看，在商用车上普及自动驾驶的意义更为重大。如果一辆卡车能够平安地从A地行驶到B地，而不需要驾驶员操作的话，就既解决了道路安全问题，又解决了驾驶员短缺的问题，更重要的是驾驶员人力成本和物流成本都将大大降低。所以我们更加关注的是SAE四级和五级自动驾驶技术，希望使卡车实现完全自动驾驶。这是一个宏伟的目标，所以也要花上不少时间才能实现。对于三级自动驾驶汽车而言，我们已经要用到一整套传感器和各种电子设备。而如果仍需要驾驶员进行操作的话，新技术带来的效益就很有限，而成本也几乎不会降低。从这一角度看，四级和五级自动驾驶技术显然更有吸力，当然也需要从长计议。

SAE：博世在这个领域做了哪些工作？

Rueger：一直以来，在乘用车领域，博世都是知名的驾驶辅助系统和自动驾驶技术提供者，但我得承认，直到两年前我们才开始真正开始研究商用车的自动驾驶。为此我们成立了一个新的团队，由我主管，试图在这一领域奋起赶上。我们正将（与乘用车）相同的技术应用到商用车上。因为所研究的系统非常先进，我们必须十分谨慎，确保所有的技术都经过严格测试并且能够真正发挥作用。商用车自动驾驶技术所使用的软件算法十分复杂，这也就意味着测试所需时间很长。如果要实现完全自动驾驶，还要有一套备用的数据传感器，并且进行传感器传输数据的融合，所涉及的范围不仅是车辆前部，而是全车覆盖。如果我们技术一下传感器的个数（参见博世的示意图），就会知道车身前后左右共有24个不同功能的传感器，这样才能很好地观察卡车周围的路况，而这还没算上拖车所使用的传感器。这是很大的一笔投入，所以也自然希望获得相应的回报。从逻辑上而言，应该先（在商用车上）使用一级自动驾驶技术，然后是二级、三级，再接下来是四级和五级自动驾驶。但如果对三级自动驾驶的投资就已经十分庞大了的话（必须同时考虑传感器及其备份装置，以及驾驶员的人力成本），我们就必须关注收益问题，看看能否实现收支平衡了。

SAE：未来的电力电子架构将会如何发展？

Rueger：电子技术是所有功能应用的基础。很显然，我们必须考虑清楚整套电子系统的架构，并且让整个架构能够满足未来七到八年的各种技术要求。如果每一家公司都仅仅埋头研发自己的技术，只顾眼前两三年而不着眼于大局，那么这些技术很可能在三年后就会遭到淘汰，然后推倒重来。考虑到如今的研发工作量之大，恐怕很难有人能够承受这样的损失。即便是乘用车来说，这也并不合适，对商用车就更不用说了。所以关于未来的电力电子架构，我们现在就要考虑清楚。这也是为什么我们要花很多精力来弄清未来的需求，并同客户一起进行了大量研究，以求在这一问题上达成共识……总的来说，我对专用型系统和封闭式架构其实并不青睐。相比之下，我更欣赏开放式的设计，因为这可以让更多人参与其中，就特定环节做针对性的研发。考虑到系统的复杂程度，我认为这恐怕是唯一正确的选择。

SAE：对采用升级改造的方式来实现汽车驾驶自动化，您怎么看？

Rueger：市场上确实有一些升级改造的方案。从系统复杂程度来看，我们已经谈过了传感器，但还需要执行机构——比如制动和转向。这是一套组合，需要电动化的转向装置来控制方向盘。我们也一直试图找出适合升级改造的方案，至少需要在理论上站得住脚。目前看来，要实现SAE五级自动驾驶，升级改造的是行不通的，因为五级自动驾驶需要（在车身周围）总共设置24个传感器，并实现电动转向，这在卡车上是无法通过升级改造实现的。在自动化程度相对较低的阶段，升级改造可能有所作为，比如实现一级范围内的列队行驶，但对真正的自动驾驶技术而言，升级改造确实并不适用，因为这对车辆效率和安全产生的影响更大。

SAE：非公路车辆方面进展如何？

Rueger：可以想见，适用于公路自动驾驶汽车的原理和技术，同样也会适用于非公路领域，因为其核心都是环境的可视化、目标物体的分类、对障碍物做出及时反应以及操作的自动化，这都是自动驾驶技术的先进之处。而当我们谈论非公路领域的自动驾驶应用时，并不仅仅是说驾驶自动化的问题，也包括作业的自动化，而从长远看，后者可能更加重要。公路车辆所使用传感技术，比如摄像头、雷达，有时还包括超声波传感器，也可以运用到非公路车辆上。比如在差车上，可以通过传感器实现俯视观察，这样即便驾驶员在搬运大型货物时无法看见前方，也能通过镜头了解作业情况。

SAE：还有没有其他已经成熟的非公路车辆自动驾驶技术？

Rueger：采用GPS导航的拖拉机或收割机并不新鲜，市场上早就有了。但GPS导航并不能清楚显示车辆周围的情况，所以如果突然出现障碍物，就需要传感器来探测了。目前我们正在研究如何减少除草剂和杀虫剂的使用，这不仅是为了环保，也是为了节省成本。这就需要用到摄像头。摄像头可用于可视化和障碍物探测，在这里则可以帮助识别小麦，甚至能辨别小麦的种类。当车子经过麦田时，喷雾器可以对准最合适的位置，可以只在探测到有小麦的地方喷洒除草剂，甚至能做到只对某种特定的小麦喷洒合适的除草剂。如果能做到这一点，除草剂成本就能降低三成到四成，这绝对是个大数目。另一个例子则是关于挖掘机动臂的辅助控制，现在市场上已经有了这样的系统，但价格昂贵，所以我们的目标就是降低成本，让该技术获得更广阔的市场。因此，不仅车辆驾驶辅助系统非常重要，作业自动化也是如此，因而也自然会涌现出各种不同的自动化解决方案。

“For Level 3, a whole redundant sensor set and electronics are already needed and the driver is still in the loop. From that perspective, Level 4 and 5 are more attractive,” said Bosch's Dr. Johannes-Joerg Rueger at SAE COMVEC 17. (image: SAE International)
Level 3 automation not attractive for trucks
2017-09-19 Ryan Gehm

In his opening keynote address at SAE COMVEC 17 on Sept. 18, Dr. Johannes-Joerg Rueger, President, Commercial Vehicle and Off-Road, Robert Bosch GmbH, noted that a fully autonomous vehicle would have approximately 100 million lines of code—about six times that for the flight software of a Boeing 787 Dreamliner. “Apparently [automation] is easier in the air than on the road,” he quipped. His point was clear—getting to fully autonomous commercial vehicles is an extremely complex undertaking.

Rueger addressed a standing room-only crowd, sharing his thoughts on the trends, benefits and solutions related to advanced driver assistance (ADAS) for on-highway vehicles and automation in both on- and off-highway operations.

Is automation more important for commercial vehicles than passenger cars?

Commercial vehicles play maybe even a more important role in automation [than passenger cars]...Analyzing data from Germany, more than 50% of accidents with casualties caused by heavy-duty trucks can be avoided with technology which currently is available—functionalities like emergency braking and lane departure warning—and another almost 40% with driver-assistance functions which will come to market in the next couple of years. [In addition to] addressing road safety, no accidents means uptime...For passenger cars it's a question of convenience; it's nice if you don't need to have your hands on the steering wheel and have leeway to do something else. But the money lies in the commercial vehicle sector. If a truck could eventually go from A to B safely without any driver, it would address road safety on the one hand and the driver shortage on the other, and particularly driver costs and logistic costs in total. So we are talking about [SAE] Level 4 and Level 5 automation, where you really have autonomous trucks. It's a big idea so it's going to take some time. For Level 3 a whole redundant sensor set and electronics are already needed; if then the driver is still in the loop, the benefit is limited and the costs are almost the same. From that perspective, Level 4 and 5 are more attractive and need to be addressed long term.

What is Bosch doing in this area?

We're a well-known supplier of driver-assistance and automation technologies in the passenger car sector, but until about two years back we didn't address commercial vehicle properly, I have to admit, and we changed that with a new organization that I'm heading. We are adapting the same functionalities to commercial vehicles...[With advanced systems], we need to be very careful that whatever is being introduced is properly tested and really works; the complexity of the software algorithms we're talking about here is really huge—that indicates that testing will take some time. If you really want to go autonomously, you need to have a redundant data sensor set and you need to have sensor data fusion, not just in the front but all around the vehicle. If you calculate the number of sensors [shown on a Bosch graphic], it's 24 different sensors for front, rear, left and right observation of the truck's surroundings, and that's even without the trailer which would need sensors there as well. So that is a significant investment. The benefit must be appropriate. If you go logically, you introduce Level 1 functionalities, then Level 2, Level 3 and then later on Level 4 and 5. But if the step to investment for Level 3 is already really significant [for the redundant layer, and the driver is still in the loop], you need to look at the benefits and whether those outweigh the costs associated with the automation.

What does the future E/E architecture look like?

Electronics is the backbone for the whole functionality. Clearly we need to think about how the structure of the electronics will look, and to build the whole architecture towards what is needed in, say, 7 to 8 years from now. If everybody develops his own little functionality by himself, and does not think about the big picture [but only] what's needed for the next 2 or 3 years, it probably will need to be thrown away in 3 years and be redeveloped. Nobody can afford that with the volumes we have. Even for passenger car, that's not really a good idea; for commercial vehicles, certainly not. It is essential to think now about the E/E architecture of the future. That's the reason why we've put a lot of effort into understanding what is needed and doing studies with our customers to come to a common understanding…In general, I'm not a big fan of proprietary systems and a closed architecture. I’m a fan of an architecture which is open in a way that gives the possibility of different parties to develop their specific content. Given the complexity we are talking about, I believe it’s the only option.

What about retrofit solutions for automated vehicles?

There are retrofit solutions in the market. If you look at the complexity, I’ve talked about sensor set but you need actuators—brakes and steering, basically—to act as well, it's a combination, and you need electrified steering to have an influence on the steering wheel. We’ve tried to indicate for which solutions, in theory at least, retrofit could be a possibility. Certainly not for Level 5 because then you have the whole sensor set—the 24 sensors [surrounding the truck]—and electric steering, and that's nothing you want to retrofit on a truck. Retrofit solutions might be a possibility for the lower-level functionalities and maybe for Level 1 platooning, but certainly not for the real autonomous driving solutions, which have a bigger impact on efficiency and safety.

What's happening on the off-road side?

Not surprisingly, the same automation principles and technologies apply for off-road operations as for on-road, because it's always about visualization of the environment, classification of objects, reactions to those obstacles, and automation, which is the distinguishing factor. When we talk about off-road operations, it is typically not just the driving which is automated but the operation of the machine itself, and potentially that's even more important. The same basic sensor set, which is camera-based, radar-based, maybe ultrasonic depending on the application, can be applied to off-road operations. Maybe for a forklift truck, sensors can be stitched together so you have a top view and the operator can easily see what's around him even if he's carrying a large good and can't really see what's in front of him.

Any other off-road applications ripe for automation?

GPS-steered tractors or harvesters—that's nothing new, they are in the market. GPS-steer doesn't control the surroundings though, so if there's an unexpected obstacle, you still need sensors to detect that. What we are working on is to reduce herbicides and pesticides use, and that is not just because we want to do something for the environment, it is a huge cost factor as well. The technology here is the use of cameras, and remember it's all about visualization and obstacle detection—in this case, wheat recognition—and maybe you can even classify what kind of wheat. When you pass over the wheat, the sprayer nozzles are in the right place and you are in a position to spray only in spots where we have detected wheat and only that kind of herbicide which is for that wheat variety. If you can do that, you can reduce costs for herbicide use by approximately 30-40%. That's really significant. Another example in earthmoving is assisted control of an excavator boom. High-priced systems are in the market already, so the target here would be to bring it on a broader scale for more affordable prices. So assistance of not just driving the machine but also of the operations is a huge topic and of course has a wide variety of different solutions.

Author: Ryan Gehm
Source: SAE Truck & Off-highway Engineering Magazine