The factory of the future full of autonomous robots is being built - BMW, NVIDIA share their progress

A backdrop of the biggest fall in car sales in decades might seem like an odd time to make significant new financial commitments, but for companies like BMW, with strong balance sheets and €14 billion in cash, COVID-19 isn’t an excuse to let strategic initiatives slide. Last week at NVIDIA’s first-ever virtual GTC event., the iconic automaker announced plans for an innovative update to its manufacturing operations using NVIDIA’s Isaac robotics platform.

As part of the company’s Industry 4.0 strategy, BMW is developing AI-enhanced robots for material handling and transport to significantly reduce the time needed to build multiple models of custom-configured cars on the same production line. According to Jürgen Maidl, BMW’s SVP of Logistics, the new system is the future of factory automation:

BMW is committed to the Power of Choice for our customers — customization of diverse features across diverse vehicles for diverse customers. Manufacturing high-quality, highly customized cars, on multiple models, with higher volume, on one factory line requires advanced computing solutions from end-to-end requires advanced computing solutions from end-to-end.

While specialized robots have long been used on automobile assembly lines, exponential increases in GPU performance with designs optimized for AI workloads enable increasingly sophisticated deep learning models that can be trained on a mix of real and synthetic 3D data. The combination results in a significant reduction in the time required to develop and train AI models to control robots capable of  handling a wider variety of parts and tasks.

Challenging times, but car makers committed to Industry 4.0 plans

The pandemic and resulting economic devastation ravaged the auto industry, cutting sales by half or more. After a challenging Q1 that saw sales drop by 20%, BMW now anticipates an operating loss in Q2 as the coronavirus lockdowns continue in the US and Europe. Nevertheless, the company remains committed to long-term strategies such as its Performance > NEXT program to considerably improve both its cost and financial performance.

Key to improving manufacturing efficiency, performance and flexibility are various robotics, AI, IoT and data analytics initiatives collectively known as Industry 4.0. McKinsey calls Industry 4.0 manufacturing’s next act and defines it this way (emphasis added):

Industry 4.0 [is] the next phase in the digitization of the manufacturing sector, driven by four disruptions: the astonishing rise in data volumes, computational power, and connectivity, especially new low-power wide-area networks; the emergence of analytics and business-intelligence capabilities; new forms of human-machine interaction such as touch interfaces and augmented-reality systems; and improvements in transferring digital instructions to the physical world, such as advanced robotics and 3-D printing.

There are four primary business goals catalyzing Industry 4.0 efforts,

1. Increase manufacturing volume and efficiency.
2. Improve product quality and consistency.
3. Reduce production costs.
4. Increase manufacturing scalability and facilitate future expansion.

MarketsandMarkets estimates consolidated revenue on Industry 4.0 technology like manufacturing AI, IoT, robotics, AR and 3D printing was about $72 billion last year. It expects annual growth of about 17 percent through 2024 that more than doubles it to at least $150 billion by 2024.

Given the benefits of increased manufacturing efficiency that lower production costs, it’s unlikely COVID-19 crisis dampens these estimates. Despite what it calls “demand vaporization in the automotive industry,” Zinnov, a global management and strategy consulting firm expects Industry 4.0 initiatives will flourish as companies are forced to become more competitive through improved productivity. One of Zinnov’s managing partners says that the post-crisis reality demands that car manufacturers focus on manufacturing and sales efficiency using automation and touchless digital experiences.

In such a challenging business environment, BMW’s investment in new NVIDIA platforms is both visionary and prudent.

The future of manufacturing logistics

BMW uses multiple categories of AI software in its manufacturing systems and incorporates the latest NVIDIA technology to make factory robots both faster and smarter using a mix of graphics, computing and software development products. BMW is currently developing five AI-enabled logistics robots in two categories:

• Logistics robots to select, grab and handle components and load carriers. These include:
  • PickBot, which takes small parts from supply racks.
  • PlaceBot, which unloads supply trains and puts boxes on the right shelf.
  • SortBot stack unloaded containers on pallets for reuse.
  • SplitBot move boxes from pallets to a conveyor belt
• Smart Transport Robots (STR) for autonomous transportation of materials throughout a factory.

In both cases, BMW uses NVIDIA’s ISAAC robotic software platform to develop various deep learning models for image perception, segmentation and human pose estimation. Traditionally, robotic control and image analysis software are trained on live data, using feedback to tweak deep learning models and narrow the gap between the algorithmic output and desired results. Recently, NVIDIA developed two simulation platforms that can greatly accelerate the process:

• Omniverse for virtual, ray-traced 3D worlds that uses real-world physics to control object movement, interactions and constraints. These virtual environments can feed data to
• Isaac Sim to simulate robots with various virtual image and depth perception (e.g. LIDAR) sensors. Isaac is used to simulate and test robot dynamics and control and sensor inputs in various environments that can include virtual human interactions.

Isaac Sim is one component of the Isaac SDK, a comprehensive development environment for robot software that also includes:

• The Isaac Engine software framework targeting applications on devices using one of NVIDIA’s edge or embedded Jetson Xavier or Nano processors.
• Isaac GEMs, which are pre-built and tested modules for frequently used functions like image segmentation, object detection, stereo/3D image processing and positioning, motor control and pose detection.
• Reference designs such as the Carter autonomous delivery robot.

The following diagram illustrates NVIDIA’s entire robot development stack. BMW feeds the real and simulated data into NVIDIA DGX systems with multiple GPUs to train and optimize deep learning models developed using the Isaac SDK.

Maidl says that such 3D modeling, simulation, visualization and deep learning systems are key to its Industry 4.0 strategy and a way to better optimize its manufacturing and logistics processes.

My take

Although GTC was unusual this year in being entirely online, NVIDIA didn’t disappoint and provided a full slate of significant updates to its deep learning processors, systems and software. These included:

• A  new GPU architecture and A100 processor with an order of magnitude performance improvement for deep learning model training and inference.
• Updated DGX systems using eight A100 GPUs and six NVSwitch fabric switches that can be clustered together to create an AI supercomputer.
• An A100-based EGX accelerator card and updated Jetson Xavier embedded system.
• A Mellanox (now part of NVIDIA) 25/50 GbE SmartNIC with an onboard processor to offload compute overhead for network encryption and virtualization.

These updates to NVIDIA’s portfolio allow companies like BMW to build more sophisticated robotic control software, reduce AI model development time and integrate multiple autonomous systems into complex manufacturing and logistics workflows. According to Christopher Brown, the co-author of a paper on transferring next-generation manufacturing and AI concepts to SMEs and cited in this piece on Industry 4.0:

New technologies such as additive manufacturing and the introduction of collaborative robots make it possible for the first time to produce even the smallest batch sizes individually and automatically. The next step in the direction of increasing efficiency and mapping a wide range of variants will be taken if robots are able to predict human movements using machine learning algorithms, thus enabling a product change in the line to be carried out on the robot, without changeover times.

BMW sees that being a leader in developing advanced robotics and incorporating autonomous machines into its manufacturing processes is critical to maintaining a competitive edge as the automotive industry undergoes wrenching changes wrought by both the coronavirus economic crash and ongoing transitions to EVs. Other manufacturers and logistics companies that expect to thrive once the economy recovers should take a lesson from BMW and not allow a tanking economy to disrupt technology strategies that will be critical to long-term success.