Quadruped robot with legs in delta configuration.

Similar projects worth following
Is it possible to build a quadruped with legs inspired in delta manipulators, using chinese 9g servos? Does it even make sense? Let's find out!


This project was heavily inspired by Simon Kalouche's GOAT concept - in which he layed out the idea of using a delta "manipulator" as a robotic leg. Check you his video - really good stuff:

His work is fantastic. If you have the time, also check his thesis work, since it's filled with juicy robotics. Here, he also sketches a robotic quadruped using his proposed leg, which was the main motivator for this tiny project of mine.  

Thanks, Simon, for coming up with this amazing idea and project! 


Well, I've always wanted to build some form of walker. I also believe that there are a lot of advantages to be found in the usage of these delta legs (as I will be calling them from now on), e.g.:

  • Omni-directionality: re-computing a gait algorithm for different heading directions becomes easier, due to the nature of the legs.
  • Reduced torque requirements for the actuators (no excessive strain on a "shoulder" servo)
  • Faster legs with reduced inertia

Those are reasons enough to try it out, I guess. :) 


  1. See if the whole thing is feasible with 9g servos instead of fancy brushless actuators
  2. Come up with a decent, omni-directional gait implementation 
  3. Learn a bit more about robotics while at it
  4. Eventually build a hexapod version of this (yeah, should have started with the hexapod, but anyways...)

  • 1 × Teensy 3.1
  • 1 × LM2596 Buck-converter
  • 1 × 16-Channel Servo Multiplexer
  • 12 × SG90 Micro Servo (9g)

  • Some detail on the mechanical side of things

    Martin Vincent Bloedorn03/03/2018 at 18:07 0 comments

      DeltaWalker has a fairly simple mechanical design, and even though the pictures do most of the talking, I'll take briefly lay out some of its aspects.

      Legs and coordinate systems

      The legs themselves were designed to be as simples as possible: 

      Three SG90 9g servo are mounted 120 degrees apart. On each leg, there are three... "Leglets", shall we say. On each, a small 3D printed hinge - the femur - is attached via the servo horn to each, and the tibia connects to the femur via a flex joint.  Since I'd go crazy trying to fit tiny bearings into 12 tiny joints, I think this flex joint is probably the most interesting thing in this whole setup. Getting the correct size and thickness for the joints was trial and error, but it actually took me two prints to settle with this design (so there's certainly room for improvement).

      The foot is also a flex, 3D printed part. There is still a lot of slippage when the DeltaWalker walks around: the foot is too low, so the plastic of the tibia ends up rubbing against the grund. This design will probably be changed in the near future. 

      Like I've mentioned earlier, I stole all the kinematics know-how from this glorious post in the Trossen Robotics forum. In that post, a few reference images are supplied (outlining parameters that are later referenced in code). Below is the one outlining the leg's coordinate frames used in the IK computation, a parameter names (conventions that I also adopted): 

      To further follow the convention, the servos are considered centered (i.e., angle = 0) when the tibia is parallel to the XY plane. Positive angles move the tibia's tip in the -Z direction.

      Body and more coordinate systems

      The whole body was designed to be printable as a single part: 

      There are some supports required for each servo attachment, but they're tiny and fairly easy to remove. Assembly is also straightforward as it seems. Just note: legs 0 and 1 are mirrored in relation to legs 2 and 3, so you'll have to mount six leglets with the femur facing one side, and another six in mirror image. 

      As a quadruped/hexapod/walker-robot newbie, I wanted to avoid going crazy with those 12 servos, so I quickly adopted a few numbering conventions, and defined some standardized coordinate frames to make my life a tad easier: 

      1. The origin {O} is placed in the body's center; +Y facing "forwards" (arbitrary). 
      2. Servos are numbered from 0 to 2 on each leg, starting from the arm, going counterclockwise.
      3. Legs are numbered clockwise, starting from the +X+Y. I actually wanted them to follow the quadrant numbering convention, but apparently I got those reversed *facepalm*. Too late to change it up though. 
      4. Each leg has a L_n coordinate system that is used by the firmware to specify the feet's positions during gait.
      5. Each leg has a D_n  coordinate system for IK computation, where Y+ is aligned with the corresponding leg's support arm (following the conventions from my IK reference). There's a Z offset a rotation between any L_n and D_n pairs, see below:

      Why so... many coordinate systems

      After I defined all these CS, it bacame really easy to manipulate the robot and do transformations with it (i.e., translations or rotations). Translations are particularly simple:

      For any desired bodily translation given by a vector d, each leg's foot has to be set to a position f = n - d, where is the foot's position when the leg is "resting", i.e., all servos are centered. 

      Yeah. That's it. Quick, painless and marginally useful :) 

  • Current state of things

    Martin Vincent Bloedorn02/25/2018 at 20:10 0 comments

    Before I go into detail on the design, assembly and firmware of the DeltaWalker, allow me to briefly outline the current state of the project.

    Although I've started it some months ago, I probably worked on it for less than two weeks on total. I started out by taking a close look at Simon Kalouche's work (mentioned in the project description), and then getting a lot of those 9g servos from China. 

    Once o had the components, I prototyped the first leg (no photos of it, unfortunately).  As you probably see in the pictures, I used 3d-printed flex parts as joints. They work suprisingly well on that scale, and I'll get into more detail on that in further posts. After a successful leg prototype, I went on to design the body and the home-milled PCB (also, more on those later). 

    The PCB essentially connects a Teensy 3.1 to a 16-channel servo multiplexer, breaks out some connections for a planned ESP8266 (and/or a bluetooth module), and allows everything to be neatly powered by a 2S LiPo battery being fed through a LM2596 buck-converter.

    Once the hardware was in place, I scraped the internet for a good implementation for delta inverse kinematics. Again, more detail on that later, but if you're in a hurry, this is gold. With those in place, I started working on a simple firmware and prototyping the gait algorithm (still wonky, as you'll see). 

    Last, but not least,  I used a IR diode/receiver pair to build a simple joystick for the robot. I'll dump IR in favor of bluetooth anytime soon, but in the meantime, it's what I had lying around.

    So, finallly, there goes a quick demo video of me playing around with some body translations and showing the (still fairly drunk) gait: 

View all 2 project logs

Enjoy this project?


Discussions wrote 02/28/2018 at 02:48 point

Awesome!! Love the concept

  Are you sure? yes | no

Martin Vincent Bloedorn wrote 02/28/2018 at 20:41 point

Thank you! Let's see how far I can go with it. Should've started with the hexapod version, though...

  Are you sure? yes | no

Morning.Star wrote 02/26/2018 at 06:24 point

Nice mechanism! Love the passive hinges, and how the legs extend... :-)

  Are you sure? yes | no

Martin Vincent Bloedorn wrote 02/26/2018 at 12:21 point

Thank you! 

  Are you sure? yes | no

peter jansen wrote 02/25/2018 at 19:54 point

Really clever!  Do the 4 primary "arms" to the delta legs have to be articulated too?

  Are you sure? yes | no

Martin Vincent Bloedorn wrote 02/25/2018 at 20:13 point

Thanks! Well, I assume they could be, but it's not the case here :) 

  Are you sure? yes | no wrote 02/28/2018 at 02:51 point

That would be next level. Delta legs with articulation would be 'leaps and bounds' over that dumb Boston Robotics dog... :P 

  Are you sure? yes | no

Martin Vincent Bloedorn wrote 02/28/2018 at 20:42 point

More joints = more fun! (if only...)

  Are you sure? yes | no

ðeshipu wrote 02/25/2018 at 18:56 point

How to make your IK calculations a nightmare ;-)

  Are you sure? yes | no

Martin Vincent Bloedorn wrote 02/25/2018 at 19:32 point

For the most part, it wasn't all that insane, haha. Thankfully, there are a lot of great resources out there! :) 

  Are you sure? yes | no

ðeshipu wrote 02/25/2018 at 21:51 point

It really makes me think about tentacles, for some reason... Short stubby tentacles.

  Are you sure? yes | no

davedarko wrote 02/25/2018 at 18:53 point

ha, never thought of that, nice!

  Are you sure? yes | no

Martin Vincent Bloedorn wrote 02/25/2018 at 19:30 point

Still figuring this out ;) Thanks! 

  Are you sure? yes | no

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates