rlmf 0.5.0

Robot Locomotion & Morphology Framework — hardware-ready legged robot API with closed-loop control

rlmf — Robot Locomotion & Morphology Framework

Universal legged-robot locomotion API. Define any robot in YAML, call robot.walk(). Works identically for 4-leg, 6-leg, 8-leg, and 12-leg morphologies.

from rlmf import Robot
from rlmf.feedback import FeedbackReader, MPU6050Sensor, FSRContactSensor

# Closed-loop (real hardware)
imu     = MPU6050Sensor(address=0x68)
contact = FSRContactSensor({"leg_0": 0, "leg_1": 1, "leg_2": 2,
                             "leg_3": 3, "leg_4": 4, "leg_5": 5})
fb      = FeedbackReader(imu, contact)

robot = Robot.load("hexapod.yaml", feedback=fb)
robot.safe_startup()
robot.walk()
robot.turn_left(45)
robot.climb()
robot.safe_shutdown()

---

Installation

pip install rlmf                   # simulation / development
pip install "rlmf[pi]"            # + Raspberry Pi hardware drivers

---

Architecture

User API            robot.walk() / robot.climb() / robot.follow_path()
      ↓
Gait Engine         TripodGait / WaveGait / RippleGait / TrotGait
      ↓
Kinematics          Full 3-D IK/FK — exact for all mount angles
      ↓
Balance Controller  PID (roll/pitch/yaw) — IMU + contact feedback
      ↓
Safety Layer        Torque · Thermal · Power · Pose drift detection
      ↓
Motor Layer         JointCalibration → TrajectoryInterpolator → driver
      ↓
Hardware            SimulatedDriver / PCA9685 / ChainedPCA9685 / custom

---

What was fixed in v0.5

IK sagittal plane error — v0.3 assumed the knee+ankle plane was vertical regardless of mount angle. For legs at ±30°/±60° this caused errors up to 180 mm. v0.5 projects the target into the leg's own sagittal frame, giving IK→FK agreement < 0.001 mm on all mount angles.

Balance controller oscillation — v0.3 used PI only. Without the derivative term, the controller overshoots and rocks the body at 1–3 Hz. v0.5 is a full PID controller using IMU gyro rate directly as the D signal.

Thermal model ODE bug — wrong denominator caused the model to saturate at ~55 °C instead of the correct ~120 °C, so thermal throttling never fired. Fixed and validated against SG90 datasheet (t_80 ≈ 90 s at stall).

Contact threshold auto-calibration — v0.3 required manual ADC threshold tuning per foot. v0.5 adds ContactCalibrator which measures rest and loaded baselines and computes per-foot thresholds automatically.

---

Morphology Description Language

name: HexaBot
body:
  segments: [thorax]
  length: 0.30
  width: 0.18
  height: 0.09
limbs:
  - type: leg
    count: 6
    joints:
      hip:   {min: -90, max:  90}
      knee:  {min: -90, max: 120}
      ankle: {min: -90, max:  90}
    segment_lengths: [0.07, 0.12, 0.10]
    calibration:
      hip:   {zero_offset: 0.0, direction: 1, speed_dps: 600}
      knee:  {zero_offset: 0.0, direction: 1, speed_dps: 600}
      ankle: {zero_offset: 0.0, direction: 1, speed_dps: 600}
physics:
  mass: 1.2kg

---

Raspberry Pi Setup

Hardware

Component	Purpose
2× PCA9685 (0x40, 0x41)	18-channel servo PWM
MPU-6050 (0x68)	IMU — roll/pitch/yaw
MCP3008 (SPI)	FSR ADC — foot contact
18× SG90 or MG996R	Joints
5 V ≥ 6 A PSU	Servo power (separate from Pi)

Wiring

Pi 3.3V  →  PCA9685 VCC (both boards), MPU-6050 VCC
Pi GND   →  common ground (all boards + PSU negative)
Pi GPIO2 →  SDA (PCA9685 A + B, MPU-6050 — parallel)
Pi GPIO3 →  SCL (PCA9685 A + B, MPU-6050 — parallel)
Pi SPI   →  MCP3008 (SCLK=GPIO11, MOSI=GPIO10, MISO=GPIO9, CE0=GPIO8)
5V PSU+  →  PCA9685 V+ (servo power rail)
Board B: solder A0 jumper → address becomes 0x41

Sequence

bash pi_demo/setup.sh              # one-time system setup
python pi_demo/calibration.py      # measure per-joint servo offsets
python pi_demo/tripod_demo.py      # walk / turn / climb demo

---

Contact calibration

from rlmf import FSRContactSensor, ContactCalibrator

sensor = FSRContactSensor({"leg_0":0,"leg_1":1,...})
cal    = ContactCalibrator(sensor)

# 1. Lift robot (all feet free)
cal.measure_rest()

# 2. Lower robot onto all feet
cal.measure_loaded()

# 3. Apply thresholds
sensor.threshold = cal.thresholds   # per-foot dict

---

CLI

rlmf robots                         # list bundled robots
rlmf describe hexapod               # topology
rlmf walk hexapod --gait wave       # simulate gait
rlmf balance quadruped              # balance state

---

Development

git clone https://github.com/Wafula-Ian01/rlmf.git
cd rlmf
pip install -e ".[dev]"
pytest -v

---

License

MIT — see LICENSE.