Imagine trying to drive your bicycle up a set of stairs — dozens of stairs. Unless you are a gifted cyclist, you likely imagined falling down and injuring yourself. Wheels work very well in a highly structured environment — on a flat surface, with defined sides for the surface you travel on and infrequent traffic intersecting your direction of travel. Folks in mechatronics would say that the environment for wheeled robots would need to be relatively controlled with similar kinds of constraints. Within these constraints, wheels are great—they’re efficient, they operate smoothly, and we have a supply chain of parts for wheels, axels, and steering components.
But stairs are less controlled environments and present a multitude of complexities. Stairs may have different heights within a single building. They may change direction, they can turn onto themselves 180 degrees, and they can be straight, elliptical, or irregular. Stairs require more spatial intelligence and kinesthetic intelligence to navigate than flat surfaces do. They pose a host of challenges for mechatronics engineers to overcome as they evolve robots for a world that goes beyond flat, defined surfaces.
In 1959, Joseph Engelberger installed his Unimate #001 robotics arm at General Motors in Trenton New Jersey to pick up diecast components. There were no sensors on this device. It had a drum memory with almost 64K of memory. It could only be safely used in highly controlled and repetitive open loop environments. “Open loop” refers to the notion that the device is not sensing, as sensing constitutes “closing the loop.” The Unimate could only perform the one task it was programmed for using its 64K of memory.
Compared to the Unimate #001, modern appliance mechatronics perform miracles in controlled environments — for example, your laundry room. Modern clothes washers sense more than the Unimate #001. They weigh the wash with strain gauges or by measuring the energy required to move the load. Dryers can sense how much moisture is evaporating off your clothes while they dry so that they know when to stop.
New sensors (e.g., strain gauges, accelerometers, energy meters, and moisture and temperature gauges) and their processors interpret data to enable our everyday appliance to close the loop. These edge digital sensors and processors can be reliable, cost-efficient and yet accurate because they sample the data infrequently, take small data samples, and use simple algorithms.
We live in highly variable environments: different stair heights and room shapes, smooth floors and carpeted floors, wet and dry, noisy and quiet. For autonomous robotic machines to just move within our environment productively, they will need the kinesthetic intelligence of a four-legged animal. To put this in context, the first vertebrate (the hagfish) is estimated to have arrived 450 million years ago. The first four-legged vertebrate was the tetrapod, 390 million years ago — so it took about 60 million years of evolution to navigate the unstructured environment of primordial earth.
Just 62 years after the introduction of the Unimate #001, Boston Dynamics created a $74,000 four-legged robot named Spot that can repetitively follow a path in a known environment. Because it is mobile it requires enough onboard processing to remember the path and the instructions for walking (remember that Wi-Fi is not always accessible). This paragon of modern technology does not yet have the kinesthetic intelligence (KI) that the tetrapod had. It can flounder on smooth surfaces, has trouble navigating around moving objects, and does still struggle a bit on stairs.
How Much Is That Doggy in the Window?
U.S. organizations spend $300 billion annually on wages for 9.1 million employees working in production-focused occupations — i.e., tasks that require repetitive heavy lifting, moving equipment, and materials. These organizations spend $15 billion on wages to complete these tasks alone (BLS, 2016), to individuals who perform their tasks with great dexterity and (for the most part) work well with their peers, their curiosity fed with new situations. The fully burdened cost to a company for each of these people is about $40,000–$60,000 a year.
Spot, at over $150,000 per year fully burdened, can only complete the most routine of tasks. And he’s always at risk of being dognapped or damaged (see HitchBot). To be a valuable partner to a human worker, robots must either exceed the performance in narrow tasks, become much cheaper to own and operate, or be able take on dangerous work that risks human lives. For example, think of a robot–human team in which the human teleoperates a robot in a dirty, dark, or dangerous (3D) environment — like the Sarcos Guardian S inspecting an asbestos plant.
Continued Intelligent Agent Evolution …
Organisms evolve largely due to conditions, stimuli, and the ability to adapt and thrive in a given environment. In the case of robotics, the key factors for evolutionary success include environmental need, technology advancement, and the ability to design a system in a way that is commercially viable.
The economics of hardware and mechatronic businesses do not improve at the rate of software or silicon. However, edge compute performance does bring decreases in costs and increases in parallelism, while deep learning brings increases in efficiency and more capabilities into the reach of commercial scenarios. As sensing AI and machine learning (e.g., object detection, pose detection, motion object identification) become commodities, these may drive interest in achieving economies of scale in mechatronics. Perhaps they will even inspire open source hardware mechatronics platforms.
Successful solutions will compete at price and performance points that that are viable in commercial markets such as retail, delivery, and security. As connectivity and edge computing evolve, the ability to connect these robots into larger networks will allow us to close business loops and achieve digital transformation.
Each cycle of automation in these markets reveals a cycle of evolution from research projects into financially successful business — and each epoch is marked by increasingly kinesthetically intelligent agents. Although I started coding when I was 14, I am more excited now than ever about how technology can make our lives more productive, safer, and more fulfilling.
You can find the original article here