• [R][T] The ideas and thoughts over the past 2+ weeks

    kelvinA • 12/07/2025 at 01:55 • 0 comments

    [Last Updated 2025 - 11 Dec]

    Inspired by the humanoid robotics industry that's been heating up in this second half of 2025, I started to formulate my own thoughts about them. After discussing with my mother, I came to a notable conclusion:

    I can last 12 days without thinking "I need to sweep".
    I cannot last 12 hours without thinking "I need to eat".

    Thus, my conclusion was that if I was ever on a cutting-edge robotics team, it would be one trying to make a robotic cook. And no I don't mean the robots that seem to be a concrete mixer food fryer, but something that can chop chives that Reddit would say are almost perfect. 

    While it's probably possible to design a fridge-freezer-airfryer-microwave-chopping-blending vending machine, the trendy way (i.e. the way to get that VC money and star-eyed engineering graduates) is via robotics.

    I was quite impressed with ALLEX from WIRobotics , particularly the robotic hands:

    Robot vacuum companies are also starting to implement arms:


    Thus, I started to focus on hands, most likely because I've already been thinking about the human hand in general due to #Tetizmol [gd0153] and #Tetent [gd0090].

    Coincidentally enough, Boston Dynamics came out with a video on their 3-digit gripper and I agree that you can do a lot with just 3. I also learned from Sunday of the importance of being able to pick up more than one object.

    Because I like symmetry, these are essentially the reason behind 6 digits.

    The next question to solve was how many degrees of freedom do they need to have, and my solution is as follows:

    • As mentioned by Boston Dynamics, the gripper has to be durable and could be fallen upon. I imagine this is why human hands can flatten out - to increase the surface area from impact.
    • There are many grasps that only need 2 fingers and no thumb, thus this concept moves them together/apart opposite but equally.
    • The fingers and thumb have 2 degrees of freedom. For the fingers, it's to be able to do things like type on a keyboard. The thumb has a rotate-out and flex range of motion.
    • The palm splits to give each hand half more space to work in. It's also to allow a single half to more easily fit into places such as a cup.

    The two fingers aren't locked to their same axis. In other words, the design should allow to "cross the fingers":

    All in all, there should be:

    • 12 DOF for the fingers/thumbs
    • 3 DOF for splitting the palm and fingers
    • 2 DOF for the wrist

    Thus 17 DOF total. It might be required that the thumbs rotate side-to-side too, requiring 19 DOF. This is the same as the recently open-sourced V2.3 DexHand:

    This hand needs to be water washable, be able to feel and have something resembling nails to be able to do things like peel an orange.

    I learned quite a bit from https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/, such as Meta's touch sensors:

    I also learned quite a bit from https://rodneybrooks.com/rodney-brooks-three-laws-of-robotics/, namely the first law:

    The visual appearance of a robot makes a promise about what it can do and how smart it is. It needs to deliver or slightly over deliver on that promise or it will not be accepted.

    From this, I can gather that the robotic hand needs 14 m/s of acceleration and cannot sound like a symphony of hobbyist servos during operation. Additionally, as mentioned in the 3rd law, there needs to be minimum 99.9%, ideally 99.999% reliability. It sounds excessive, but if not falling was 99.9% reliable, people would trip and fall every thousand steps on average. This means characterising wear-and-tear.

    Obviously, having the idea and implementing the solution are vastly different things. I just wanted to write this for the potential chance that someone on a team to design yet another robot arm stumbles upon it.