Acromonk

Build the test environment into the plan

The hardware guide describes six unique printed structural parts, a Raspberry Pi 4 Model B, mjbots pi3hat, a qdd100 beta 3 servo and a 6S 1200 mAh LiPo battery. The archive includes an assembly STEP file, and the guide supplies a wiring diagram and a remote emergency-stop circuit. The ladder-bar rig is part of the experiment: upstream calls for a cage, access to the stop and at least one metre of separation during operation. Those requirements matter when deciding whether a lab can accommodate the platform. [S001] [S005] [S006] [S013]

The BOM reports an estimated total of 950–1000 under a Germany/euro heading, but two major components are explicitly priced in dollars. It also excludes shipping and printing. Without a stated currency conversion and a current configuration-specific quote, that figure cannot serve as a verified delivered-build budget. This catalog leaves the current complete cost unknown. [S005]

Begin with the historical simulation environment

The simulation guide reports testing on Ubuntu 20.04 and 22.04. It uses Drake for trajectory work, while the reinforcement-learning path uses MuJoCo 2.1 through mujoco-py; the requirements also name older NumPy, SciPy and Stable Baselines versions. This is a specific historical environment, not evidence that installing the newest Drake or native MuJoCo package will reproduce it. No dependency installation was performed for this review. [S007] [S009]

The documented trajectory workflow chooses an atomic behavior, generates a solution with direct collocation and saves trajectories and hyperparameters. A separate stabilization script evaluates TVLQR or PID gains and writes closed-loop outputs. That provides a useful first research task before attempting motion on hardware. The repository also contains a trained-controller archive, but its contents and training provenance were not validated here. [S007] [S013]

Initial configuration changes the interpretation of motion

The physical-system guide assumes execution on the Raspberry Pi and has its own moteus and moteus-pi3hat dependencies. Motor ID, CAN bus, trajectory and recovery files are configurable. The motor zero is established when power is applied with the arm hanging at the documented zero angle. INIT_CONF changes whether the IMU arm swings or stays attached and changes the motor-axis sign convention. These are commissioning inputs, not optional presentation settings. [S008] [S010]

The Makefile passes controller and configuration arguments to the hardware script, and the script reads the requested brachiation count from the NBRACH environment variable, defaulting to one. The saved script does not implement the guide’s complete execution path: its kickback, swing and catch wrappers return early, and generate_aa returns a constant instead of checking arm attachment. It also calls go_to_pos before displaying the “Start?” prompt. A lab must reconcile the guide with a matching implementation before planning hardware reproduction; removing those returns alone is not a validated repair. This review did not energize an actuator or execute the script. [S008] [S011] [S012]

What the archive establishes

The project declares its work BSD-3-Clause and explicitly separates third-party software terms. CAD, integration documentation and control code are present. That does not establish the licenses of every purchased board or external firmware dependency, nor the provenance of the supplied trained-controller file. The guide/code conflict is an observed limitation of this saved revision, separate from the absence of testing. Manufacturing quality, simulation fidelity and repeated brachiation have not been tested by this publication. [S001] [S003] [S006] [S008] [S012] [S013]