TL;DR — Key Takeaways

  • Hands Could Be Robotics’ Biggest Advantage: Humanlike dexterity could allow robots to operate existing tools, open doors and handle unfamiliar objects without expensive changes to workplaces.

  • Simple Grippers Still Get the Job Done: Agility Robotics’ Digit demonstrates the commercial value of specialized robots that reliably perform repetitive tasks without needing humanlike hands.

  • Forget the Demos. Show the ROI: The real test of a general-purpose robot isn’t whether it can play the piano, but how many useful tasks it can complete without human intervention.

Some robots walk, some robots rock, some robots roll. Agility Robotics’ Digit walks on two legs and moves totes in a warehouse. Genesis AI’s Eno rolls on wheels and reaches for a box with hands modeled on ours. Eno can also play the piano, at least in a company demo. I’ll give it credit for the piano. If I were buying one for a warehouse, I’d ask what else it could do with those same hands between the start of a shift and the end of it.

Genesis just raised $500 million at a $3.5 billion valuation. Agility and logistics operator GXO says Digit has already moved more than 100,000 totes at one Georgia facility. The Genesis round tells us how much investors are willing to put behind a robot that could do many different jobs. The tote count tells us how much work a robot can do when its job is clearly defined. Put Eno next to Digit and the design choices are almost reversed. Eno has wheels and humanlike hands. Digit has legs and simple grippers. Both borrow from us, and each borrows a different part.

Think about coming home with a bag of groceries. One hand holds the bag while the other fishes a key out of your pocket. You recognize the key by feel, move it into position between your fingers, slide it into the lock and turn it. If it catches, you ease it back a little and try again. Then you change your grip to open the door. You can do most of this while talking to someone or thinking about what you forgot to buy. Ask an engineer to build a machine that can do it with the same hardware it uses to pack the groceries and you begin to appreciate what evolution gave us.

The other half of that advantage is ours to take credit for. We made the key, the lock, the bag and the door handle for hands like ours. We did the same with screwdrivers, kitchen drawers, valves and controls on factory equipment. A robot that can use those things can go to work in places people already work without someone first replacing the handles or buying a new set of tools. That’s an engineering consideration with a dollar figure attached. How much of the building or the process does the customer have to change before the robot can do the job?

What a Gripper Does Well

There’s a reason factories use grippers. Give a two-jaw gripper a part that arrives in a predictable orientation and it can pick it up firmly and put it down accurately over and over. Some grippers adapt to different shapes or let the operator dial in how hard they squeeze. Robotiq’s 2F-85, for example, specifies a range of 20 to 235 newtons of grasp force. You can get a lot of useful work out of two fingers. Digit’s tote count is a reminder of how easy it is to underrate the value of doing one job reliably.

Where the hand earns its keep is after the pickup. Can the robot turn a part over within its grasp so it can inspect the other side? Can it feel a slippery object starting to move and tighten its grip just enough to keep it from falling? Can it hold a tool and work the trigger with one finger? A robot can sometimes accomplish the same thing by moving its wrist, using a fixture or setting the part down and picking it up again. On a dedicated production line, that may be an excellent solution. When the work changes several times a shift, each extra fixture, tool change or trip back to the bench starts to count.

Boston Dynamics and China’s Unitree show how much room there is between a clamp and a replica of a human hand. Atlas uses two fingers and a thumb. Boston Dynamics has talked about the trade-off between dexterity and building a gripper that can handle heavy objects and survive a collision with a table. Unitree’s G1 can use its Dex3-1 hand, also with three digits, but with seven degrees of freedom and tactile sensing. Counting fingers would put these designs in the same category. The way those digits move, what they can feel and what they can withstand tell you far more about the work each might do.

Figure and 1X are putting more of their design effort into the hand. Figure says its Figure 03 fingertip sensors can register a force equivalent to the weight of a paper clip. The point of that sensitivity becomes clearer if you imagine picking up a paper cup without crushing it or noticing when it starts to slide. 1X has developed tendon-driven hands for its NEO humanoid, borrowing another feature of human anatomy. Those are interesting design choices. I’d want to see how either robot handles a drawer full of unfamiliar objects, and how often somebody has to help it finish.

China’s Robots Make Different Bets, Too

Unitree is only one example. Galbot’s G1, which shares a model name with Unitree’s robot, takes a route closer to Eno’s. It puts two articulated arms on a four-wheel base that can move in any direction. Picture it sidling up to a shelf on a level floor. Now picture the same shelf at the top of a staircase. Where the work is located tells you more about whether to buy wheels or legs than a video of a robot walking across a stage.

Fourier takes touch in a different direction. Its GR-3 offers articulated hands and puts touch sensors on the head and torso as well. A hand needs to register contact to manage a grasp. A robot working near people can also benefit from knowing it has made contact somewhere else on its body. UBTECH’s Walker S2 brings another priority into the picture. UBTECH says the industrial biped can swap its own battery in three minutes. For the person responsible for keeping a line running through the second shift, that feature may get more attention than anything the robot can do at a piano.

What the Buyer Should Ask

If I were looking at these machines for a factory or a warehouse, I’d start with the jobs and the building. How much of the shift consists of moving the same item between the same two points? How many jobs involve opening something, using a tool or handling parts that come in different shapes? Can one robot cover enough of those jobs to justify what it costs to buy, maintain and repair? I’d ask the vendor to run it on my floor with my parts, then count how many times a person has to step in. I’d want those numbers after the demo team has gone home.

Genesis is also betting that a humanlike hand can make teaching a robot easier. It designed a glove to capture the movements of a person’s hand and a robotic hand that closely matches it. A worker who knows how to do a job can demonstrate it in a way the robot’s training system can capture. That’s useful if the robot can then handle the variation the worker sees every day. Watch someone who has done assembly work for years, and you’ll see the little adjustments they make when a part comes in slightly crooked. I’d want to know how much of that judgment carries over along with the motions.

I expect plenty of robots to earn their keep with wheels and grippers. A dedicated machine that moves totes all day can free up people for other work, and the case for it is easy to understand. The more interesting bet is what happens when we ask that same machine to leave the loading area, get through a door and do something it wasn’t doing five minutes ago. There’s value in taking a machine to the work as it already exists, especially when changing the work would mean modifying thousands of rooms or replacing tools people have used for years.

The next time you get home with a bag of groceries, pay attention to what your free hand is doing. It finds the key by touch, shifts it into position and feels when it’s seated in the lock. By the time the door opens, your fingers have made a series of adjustments you probably couldn’t describe without doing it again. There’s plenty of room for engineers to rethink the rest of the robot. For the part that gets the key into the lock while you’re still holding the groceries, I’d give evolution a tip of the hat.

Frequently Asked Questions

Why are robotics companies developing humanlike hands?
Humanlike hands could enable robots to manipulate tools, operate door handles and handle irregular objects in environments originally designed for people. This flexibility may reduce the need to redesign workplaces around robotic limitations.
Are dexterous robot hands better than traditional grippers?
Not necessarily. Traditional grippers are often simpler, more durable and highly effective for repetitive industrial tasks. Dexterous hands offer greater flexibility for manipulating unfamiliar objects, using tools and performing varied work.
What should businesses consider before investing in humanoid robots?
Businesses should evaluate the tasks a robot can complete, its reliability, maintenance costs, required infrastructure changes and frequency of human intervention. Real-world performance and return on investment matter more than impressive demonstrations.

TECHSTRONG AI PODCAST

SHARE THIS STORY