🕷️ SpiderBot

A 12-DOF Quadrupedal Robot Daniele Palomba & Giulio Vestri — Final Project for Advanced Robotics, University of Naples Federico II.

---

🚀 Project Overview

SpiderBot is a quadrupedal robot with 12 degrees of freedom (DOF). The physical architecture is an evolution of an original hexapod design created by user emrekalem. We structurally modified and adapted the original files into a quadrupedal configuration.

The project covers 3 different macroareas: from open-loop simulation and training via end-to-end Deep Reinforcement Learning (DRL), through to physical implementation on real hardware.

✨ Key Features

• Dual Control Paradigm:
  • Open-Loop Kinematics: Time-varying analytical gait generator (Creep, Trot, Gallop, Pace).
  • End-to-End Neural Controller: Control based on DRL policies learned directly in accelerated physics simulation.
• Efficient Hardware: Master-Slave architecture based on an Arduino MCU and PCA9685 PWM driver via I2C for synchronous management of 12 MZ996R servomotors.

---

🛠️ Control Architecture

1. Open-Loop Kinematic Control

The classical kinematic pipeline, implemented in MATLAB/Simulink, is structured on three hierarchical levels:

1. Gait Generator: Computes the desired 3D Cartesian coordinates of each foot relative to the robot body frame.
2. Inverse Kinematics: Converts Cartesian footprints into joint angles via a geometric Inverse Kinematics (IK) solver.
3. URDF / SpiderBot: Sends PWM commands to individual servomotors (Coxa, Femur, Tibia) while synchronizing the digital twin.

🐾 Implemented Gaits

Gait	Type	Foot Scheduling Parameters	Description
Creep Gait	4-beat	[0, 25, 50, 75] (T = 2.0s)	Maximum stability on uneven terrain.
Trot Gait	2-beat	[0, 50, 50, 0] (T = 0.6s)	Medium-to-high speed gait with synchronized diagonal pairs.
Gallop Gait	High speed	[0, 10, 50, 60] (T = 0.6s)	Maximum dynamic reactivity.
Pace Gait	2-beat	[0, 50, 0, 50] (T = 0.8s)	Coordinated movement of limbs on the same side.

2. Deep Reinforcement Learning (End-to-End)

As an alternative to classical control, SpiderBot implements an End-to-End Neural Controller paradigm. The neural network maps the physical state directly to joint angle targets.

• Observation Space (35D): Includes orientation (quaternions), angular velocity, gait phase, and current joint positions.
• Action Space (12D): Angular targets in $[-1.5, 1.5]$ rad for the 12 motors.
• Teleoperated Learning Reward: Multi-axis tracking optimization via exponential Gaussian reward functions, with dedicated penalties to eliminate lateral drifts.

---

💻 Repository Structure

├── Matlab/              # Simulink open-loop models and related graphs
├── MuJoCo-Jax/          # DRL training pipeline, enjoy_env, export_json, policies, and XML model
├── Mujoco-Jax-TEL/      # Telemetry, enjoy_env, export_json scripts, policies, and XML model
├── Spiderbot/           # Arduino firmware for the physical robot and MATLAB script for teleoperation
└── Spiderbot_urdf/      # URDF representation of the robot

---

🔌 Electronics & Hardware Architecture

The physical system is based on a Master-Slave topology to optimize computation cycles:

• External Host (Master): Handles high-level command streaming or execution of the neural policy.
• Local MCU (Slave — Arduino): Receives serial data, computes gait phases, and performs Inverse Kinematics geometry in real time.
• PCA9685 PWM Driver: Dedicated I2C actuator that offloads MCU timers and drives the 12 MZ996R servomotors with precise 12-bit pulses.
• Chassis: Entirely 3D-printed using FDM technology (PLA), matching the kinematic dimensions of the URDF model down to the millimeter.

---

🚀 Getting Started

1. Kinematic Simulation (MATLAB)

1. Open MATLAB and navigate to the MatLab/ folder.
2. Open and launch the Simulink model open_loop.slx with chosen gait.
3. Run graphs.m to generate all the necessary graphs.

2. DRL Training (Python/JAX)

Set up your Python environment and install the required dependencies:

cd MuJoCo-Jax
pip install -r requirements.txt
python train_spider.py

To visualize results or deploy a pre-trained policy:

python export_json.py 
python enjoy_spider.py 

---

Project developed in collaboration with the PRISMA Lab.