Deploying Artificial Intelligence where the ‘Rubber meets the Road’

A long-haul truck cruises by you on Interstate 10 in west Texas. You get a brief wave from the driver who seems unfocused on the road ahead.  It’s hot and the road is bumpy. You didn’t realize it but you just encountered an autonomous driving truck in development, testing out the latest version of its Artificial Intelligence algorithms.  The truck is loaded with unseen video cameras, lidar, radar and infer-red sensors.  As it travels along it is seeing its environment and its on-board computers are making the thousands of little decisions that keep it moving safely to its prescribed destination.  The driver is a safety layer that will be removed in future iterations of the design.

Meanwhile flying over a remote desert, a reconnaissance aircraft is gobbling up data at an amazing rate and using its artificial intelligence capability to build a sophisticated understanding of the environment it is operating in, both the natural and manmade, determining friend and foe.  The plane is working in conjunction with a mobile command center below which is aggregating the intelligence from scores of sources and applying its own artificial intelligence to inform real time tactical decision making on the ground.

The truck is operating in extremes of hot and cold with constant shaking in an environment of dust and humidity with limited power available for computer equipment. In the plane the vibration is constant, with the occasional shock of unexpected turbulence, the altitude effects the ambient temperature, and space and weight are at a premium.  The power on the plane is limited and unique. On the ground the mobile command center has its own environmental challenges with shock and vibration while in transit and the need to continue operations in extreme hot and cold temperatures. These are not environments friendly to sensitive electronic equipment.

These applications are like many similar in autonomous vehicles in mining, construction, or agriculture or in military applications on or under the sea.  They share the common trait of needing to deploy the most sophisticated Artificial Intelligence but demanding that that capability be delivered in unforgiving harsh environments.  These are environments where your typical computer systems cannot operate.  This is the emerging challenge; how do you deploy the most capable AI platforms where the rubber meets the road.

The first thing to understand is that these applications will benefit from the very highest level of AI performance.  The more performance available the more data can be utilized and the more insightful AI conclusions can be obtained.  This means the AI platforms need to be no-compromise in the level of hardware technology they include; the latest and most power CPU, GPUs, co-processors, memory, and storage. Today GPUs are the workhorse of AI computation.  High end GPUs utilize 100’s of watts of power and a system can require multiple 1000’s of watts.  The electronics in these systems generate a significant amount of heat exasperating the need to cool the system in harsh conditions.

Unlike in the controlled environment of a datacenter with constant cool ambient temperatures, stable and abundant power and no need to accommodate for shock, vibration, humidity, or dirt these AI platforms must be built to tough standards.

The platforms must first be rugged.  Their mechanical design and material selection needs to accommodate the instability of mobile deployment.  They must be compact with form factors and mounting strategies that accommodate the unique space available on size optimized vehicles.  They must include innovative cooling design that deals with extremes in ambient temperature while cooling powerful hot components.  Strategic air-cooled topologies or the use of liquid cooling is required. Additionally, vehicles on land, sea or air will have different power sources available ranging from terrestrial vehicle providing 48V DC while some aircraft supply 3 phase 400Hz power.

With the proper design all of these requirements and constraints can be met but not with your standard off-the shelf HPC systems.  What is required is a modular standard edge supercomputer that can deliver datacenter class AI performance with the rugged flexible design to accommodate harsh operating environments.

Now when you are passed by that long haul truck testing its autonomous capabilities you can think about the engineering challenges that need to bead dressed to provide for its ever-expanding intelligence requirements.  And give that driver a wave, you might not see him next time.