This robot is inspired by Barker Redbacks FTC robot, we saw their super light arm and wanted to adapt it to work for FRC. Once we had started the design process, we realized that we could use a system of zombie axels (axels that are both dead and alive) to move all the motors to the base of the robot. While this is not a complete robot design the Zombie Arm is fully functional and complete.
We had a few main goals for this robot when we started to design it. The main ones were having it super light, having a low center of gravity even when fully extended, and having no wires or motors visible on the robot.
So what is a zombie arm?
It’s an arm that uses zombie axles which allows all motors to be placed at the base of the arm instead of being on the arm.
Why do this?
Keeps the arm lighter which keeps the CoG lower and makes it easier for the arm to be moved.
All motors are centralized so wiring can be neat or hidden easily.
Why stop at just two degrees of freedom?
Adding on a wrist or other degrees of freedom would be easy by running more belts/chains up the arm.
Robot stats
55.8 lbs without end effector (didn’t design one because that is too game-specific)
22.5 square drive base
4.52 arm weight without end effector
Key features
Absolute encoders on both major joints allow for easy setup during matches
Simple chain tensioning for both shoulder and elbow joints
All electronics and motors beneath the top surface of bumpers
While this design seems very promising there are some limitations. This would only work for light-ish game pieces as the arm is not as robust as other designs. This design works best when you only hold one game piece at a time (eg. 2011, 2019, 2023). This design does have many advantages that would make it very useful in some games, it would be simple to make spares for, easy to set up, and simple to do the required maintenance like chain tensioning.
971’s arm used a deadaxle for the base pivot bc 1/2 hex pivot even on a zombie axle will die under bending, however. The other major concern with putting motors as the base is backlash, which is extremely bad and often can make systems uncontrollable. Arms are very sensitive to backlash, which is often why we rarely see double-jointed arms built this way, its often better to just take the worse CG for lower backlash unless you like wanna use steel cable.
I also helped design this robot. While I did see 971’s robot we believed that it was different enough as the main goal with this design was to move the motors on the end effector lower not just the ones for the elbow. We were not trying to claim that this is the first arm to have the joint motors on the base but, from our research it’s the first one with all the motors, including for the end effector, on the base. And while backlash is a slight concern, the lighter arm will make reducing backlash easier especially with the chain tensioners.
Just pointing this out, but on your design the encoders aren’t doing anything. The rotation of each arm joint is isolated with bearings. If you attached the arms with a hex hub instead of a bearing it would be a zombie axle.
I am not sure I would take this as a generalization. (But I do see what umbrella concept you are getting at)
We were fighting for low COG on 2023 for a variety of reasons, setting rigidity aside for a moment some mass did prove handy in the system dynamics to dampen some natural frequencies.
As for the design I would personally go for some large diameter capstan drives with dyneema for backlash minimization (and availability/cost). 2451 ran mostly this for the shoulder and elbow (wrist and gripper on neos at end of the arm). However this is highly game dependent.
There is one cool thing you can do if there are no wires up the arm… You can make it pass through itself infinitely if desired. (Maybe use a slip ring for sensors if you are ambitious, control a degree of freedom or two with a differential if more ambitious)
Great job, 8724 folk! Double-jointed fixed-motor arm architecture is a great choice, it’s very common in the Real World:
Keep working on it! Some things you might want to think about for your next iteration:
absolute encoders on the joints may not give you the spatial resolution you want: their resolution is not infiinite, and the noise is not zero, and the flex in the system isn’t zero either.
you’ll probably want to (greatly) increase the diameter of the drive components. the larger they are, the lower the stress on the pulling-and-pushing parts (e.g. the chain) and the lower the backlash. for comparison, the 971 drive sprocket in 2023 was about eight inches in diameter.
you may want to consider a four-bar strut (as in the ABB design) rather than a chain, since it can be made simpler, lighter, and lower-lash.
consider (greatly) increasing the depth of the “beams” that make up the arm. stiffness is the cube of depth, so it makes an enormous difference. the parallel-narrow-tube design you have now will be very flexible. try a single beam for each segment with a four-inch cross-section, see how light you can make it.
Think about back-drive and lash in the four-stage drives you’ve designed. How could you reduce the number of stages to, say, two?
Our team did something very similar in 2023. It was quite nightmarish but I do believe that is mainly due to poor implementation and manufacturing. It was quite hard to control.
This has always been possible, but people have decided not to because backlash is just that bad where most other solutions are preferable. The easiest and most effective way to remove backlash to to reduce or eliminate adapter stack, which you can’t do with the motors at the base due to numerous adapter stack, as hex/hex interfaces are just terrible in terms of backlash due to imperfect tolerances and wear making those tolerances worse.
Combining all the parts into fewer rigid parts saves adapter stack, but is heavy the point where you’d rather just throw a brushless motor on the end. You still have to deal with the compliance of belts and chain even if you eliminate all hex/hex interfacnes, which is not 0, and w
steel cable is better in this regard, they have a limited range of motion which won’t work for the “everything at the base” approach.
All else being equal, a stiff but heavy system will outperform a system that’s light but with stiffness or backlash issues because the latter doesn’t care about your driver skill or controls, it will do what inertia wants.
Even if it was manufactured better, double joined arm is still inheritly hard to control due to being coupled non-linear systems. There are ways of doing it, but they are significantly harder that PID + feedforwards. Both the manufacturing problem and the controls problem have to be solved to make a nice double-joined arm, hence the rarity of good double-jointed arms in FRC.
I’d have a look at the redbacks 2023 double arm, very similar in a lot of ways to their 2025 ftc arm. Double jointed, carbon fiber, powered by belts with motors at the base.
I have not personally programmed one, but I have to imagine there are a few setpoints which the driver can command the arm to go to with closed loop control and a single button push. If small adjustments are needed then maybe a manual adjustment joystick in addition? But I think for most FRC games a few setpoints mapped to buttons should be enough.
Thank you for sharing this, we didn’t find them when we did our research. This looks like the same idea as what we came up with, so we aren’t the first . But it looks like it worked really well for them
The current plan that we have for if we were to program it is using a set of preprogrammed positions mapped to some buttons. The travel in between positions would be controlled through inverse kinematics, and manual control would be using a joystick to change the x,y coordinates of the end effector. None of this has been done yet, this is just our plan, so if there are any major issues with doing it like this please let us know.
.
. But it looks like it worked really well for them