⚙️ Hardware Specifications

Under the Hood

Every sensor, motor, and microcontroller that makes GRACE tick. A distributed embedded system with AI muscle.

The Robot

Meet the Hardware

GRACE physical robot base - real hardware with LiDAR, ultrasonic sensors, and hoverboard wheels

The Physical Base

GRACE's base houses the hoverboard BLDC motors, LiDAR, ultrasonic sensors, battery packs, motor drivers, and the main electronics. The transparent top cover shows the internal wiring and PCBs.

  • Hoverboard BLDC motors (differential drive)
  • RPLidar A2M7 mounted on top
  • Ultrasonic sensors on front face
  • Red power switch + safety kill
  • Transparent acrylic top cover
  • Tested with 84 kg load for ~2 hours
GRACE base CAD design in SolidWorks - circular base with hoverboard wheel and LiDAR mount

SolidWorks CAD Design

The base was designed in SolidWorks with DFA + DFM optimization. Circular aluminum base plate, integrated motor mounts, and sensor housings — all designed for manufacturability.

  • Circular base plate design
  • Integrated LiDAR mount (center top)
  • Hoverboard wheel on each side
  • Ultrasonic sensor ports on front
  • Internal compartments for electronics
Compute

🧠 Core Brain

A dual-processor distributed nervous system — heavy AI on the Jetson, real-time sensor I/O on the STM32.

🟢

Jetson Orin Nano Super 8GB

The main AI brain. Runs all heavy computation:

  • Computer vision & pose detection
  • SLAM (Simultaneous Localization and Mapping)
  • Nav2 autonomous navigation
  • Contextual reasoning & AI inference
  • ROS 2 Humble master node
  • JetPack 6 (Ubuntu 22.04 ARM64)
🔵

STM32L475 (B-L475E-IOT01A)

The sensor aggregation MCU, communicating via micro-ROS:

  • Collects all environmental sensor data
  • Battery voltage & current monitoring
  • Temperature, humidity, pressure readings
  • Real-time I/O at microsecond latency
  • micro-ROS agent → ROS 2 bridge
Sensors

👀 Perception Stack

📡

RPLidar A2M7

  • 360° 2D laser scanner
  • 12m range, 8000 samples/sec
  • Used for SLAM + real-time navigation
  • Mounted just above wheel plane
📷

Intel RealSense D435i

  • Stereo RGB-D camera
  • Built-in IMU for motion tracking
  • Depth range: 0.2m – 10m
  • Human detection & following
🎥

RGB Camera

  • General-purpose vision
  • Pose detection via Jetson GPU
  • Context-aware AI inference
  • Publishing on ROS 2 image topic
🌡️

Possible Thermal Camera

  • Thermal-based anomaly detection
  • Fever detection at distance
  • Night-time monitoring
  • Under evaluation for integration
🔊

Ultrasonic Sensor

  • Close-range obstacle detection
  • Safety fallback for LiDAR blind spots
  • Works on transparent surfaces
📐

3× IMU Sensors

  • Main IMU on base_link
  • Front IMU for tilt detection
  • Back IMU for stability monitoring
  • Fused with EKF for localization
Environment

🌍 Environmental Sensors

🌡️

Climate Sensors

  • Temperature sensing
  • Humidity measurement
  • Barometric pressure
  • Indoor comfort monitoring
💨

Air Quality

  • CO₂ concentration
  • CO (carbon monoxide) detection
  • PMS5003 particulate matter sensor
  • PM2.5 & PM10 levels
⌚

Health Wearable

  • Smart wristband integration
  • SpO2 (blood oxygen) tracking
  • Heart rate monitoring
  • Lifestyle metrics & activity
Drive

🚗 Locomotion System

GRACE locomotion - SolidWorks view showing hoverboard wheel

Hoverboard BLDC Motors

GRACE uses repurposed hoverboard brushless DC motors for a differential drive configuration. Robust, high-torque, and energy-efficient.

  • Differential drive (2 powered wheels)
  • 4 passive caster wheels for stability
  • 5-bit magnetic encoders per wheel
  • Wheel separation: 454.68 mm
  • Wheel radius: 82.55 mm
  • Max linear velocity: 0.7 m/s
  • Max angular velocity: 8.5 rad/s
  • Separate battery system for drive vs electronics
0.7
m/s Max Speed
8.5
rad/s Max Turn
454.68
mm Wheel Sep.
82.55
mm Wheel Radius
Structure

🏗️ Mechanical Design

GRACE physical robot chassis

Weldment-Based Steel Frame

Designed for manufacturability (DFA + DFM optimized), not just academic demonstration. A practical, production-oriented chassis.

  • 20×20 mm mild steel square pipe frame
  • Weldment-based construction
  • Bottom-heavy layout (motors, batteries, drivers below)
  • LiDAR mounted just above wheel plane
  • Cameras & sensors mounted higher
  • Circular aluminum base plate
  • DFA + DFM optimization in progress
Software

💻 Software Architecture

🤖

ROS 2 Humble

  • Full robot middleware
  • DDS-based communication
  • diff_drive_controller
  • twist_mux for cmd_vel arbitration
  • robot_state_publisher + TF2
🗺️

Nav2 + SLAM Toolbox

  • Autonomous navigation stack
  • Online async SLAM
  • DWB local planner
  • NavFn / Smac global planner
  • Behavior trees for recovery
📍

Localization (EKF)

  • robot_localization package
  • IMU + wheel odometry fusion
  • Extended Kalman Filter
  • Reduces slip-based drift
🎮

Control & Safety

  • twist_mux priority-based arbitration
  • Joystick: priority 99
  • Web control: priority 95
  • Nav2: priority 80
  • Safety stop lock: priority 255