After CADing up the second revision of the chassis, I set to work with the 3d printer and printed up all the pieces.

There were a few minor post-modifications I had to make, which were all much faster than printing the pieces again.  All the holes for M3 bolts were slightly undersized, so I drilled them out.  The battery holder had a channel to let the power wires out, which inexplicably terminated before reaching the edge of the holder.  I also had to install all the heat set inserts.

As described in my roadmap, the chassis for the quad A0 was on the verge of failing, or causing the shoulder motors themselves to fail, after only a few hours of walking around.  Also, it was nigh impossible to assemble, disassemble, or change anything about it.  Thus, the chassis v2!

More than one piece

The old chassis was a single monolithic print that took about 35 hours of print time.  Because of its monolithic nature, there were lots of interference problems during assembly.  For instance, the shoulder motors could only have 4 of the 6 possible bolts installed, and 2 more of the bolts extended beyond the chassis entirely.  I decided to break it up into multiple pieces, which uses a lot more inserts and bolts, but should allow for a feasible order of assembly and manageable repair.

Now that I have a full rate inverse kinematics and dynamics solution, I can begin to do more interesting things.  A while ago I did the first jump on the quad A0 – in that video I used a limited technique just to verify that the platform was indeed capable of jumping.  The joints were commanded in an open loop fashion, and really only at the transition points of the jump sequence, relying on the control loops in the servo to actually achieve each stage of the jump cycle.  That resulted in the jump only being minimally controlled… tracking errors would result in the robot taking off from a not-level position and the timing was not super reliable to boot.

Last time I updated my inverse kinematics solution to also include dynamics, velocities and forces.  Now I’m in the process of integrating this onto the robot.

The old SMMB / HerkuleX control software commanded the servo positions in an open loop, which did not take into account the actual position of the joints in any way.  What I’ve done now is implemented a control flow where at each cycle the state of all 12 servos is sampled, then the control laws are applied based on that state, then the resulting commands are sent out.

A while ago I derived some closed form solutions for the inverse force dynamics for a 2 dimensional mammal geometry leg.  Now that I want to execute more dynamic gaits, I need to be able to control the velocity and force of the 3 dimensional leg in real time.  That means I need to be able to go forward and back between the two representations.  The first being joint angles, joint velocities, and joint torques – the second being 3D position, velocity, and force.

A previous simulator I had built for Super Mega Microbot was based on the “DART” robotics toolkit.  It is a C++ library with python bindings that includes kinematics, dynamics, and graphical rendering capabilities under a BSD license.  I wanted to use some of its dynamics capabilities for future gait work on the quad A0, and eventually re-incorporate its simulation capabilities, so integrated a subset of it into mjbots/bazel_deps.

• github: Incorporate a subset of dart and its dependencies

My outdoor filming for the project update video was cut short when the machine cut power to the motors, fell down, and one of the legs snapped off.  Fortunately, I already had plenty of footage when that happened, so it didn’t really impact the video.

First, this demonstrates the not too surprising fact that this particular part of the leg design could use to be improved.  Second, and the topic of this post, is improving what the machine does when the inevitable failure does occur.

To verify that I could make FD-CAN work in the next revision of the moteus controller, I made a simple desk setup between two NUCLEO-G474RE boards.  I started by soldering up some breakout boards for the TCAN334G CAN transceiver I’m planning on using:

And then wired those up with a lot of jumper wires:

After a fair amount of fiddling, bisecting against the ST CUBE example project, and fixing some problems with my STM32G4 support in rules_mbed, I ended up with some 16 byte CAN frames being sent and received with a data rate of ~4Mbit.

While working on the next revision of the moteus controller, I started by bringing up a software toolchain on a NUCLEO-G474RE board.  Unfortunately, even the most recent mbed 5.14 doesn’t yet support that processor.  Thus, I have a half-baked solution which pulls in the ST CUBE sources for that controller into the mbed tree as a patch.

https://github.com/mjbots/rules_mbed/blob/master/tools/workspace/mbed/stm32g4.patch

The patch only implements wrappers for the things I care about, so it isn’t a complete solution. Since I am not really using any mbed libraries anymore in any project, that isn’t a whole lot.  Right now I’m just using it for the one function that sets up the board, a linker script, and the pin mappings.  I will probably eventually just make a rules_stm32 and ditch mbed entirely but for now that is more work than it is worth.

While bringing up an STM32G4 for the new revision of the moteus controller, I wanted to be able to flash and debug the system, and thus needed a working OpenOCD installation.  The NUCLEO-G474RE board has an ST-LINK-V3 debug interface, which no released version of OpenOCD supports, although thankfully that is working just fine at HEAD in git.  However, to make the STM32G4 work I had to pull some patches from the sysprogs/openocd tree.