An advanced autonomous mobile robot platform equipped with ROS integration, vision systems, and real-time control for autonomous navigation research and indoor robotics applications.

Project Overview

Hykabaï is an advanced autonomous mobile robot platform engineered for research and autonomous navigation applications. The system integrates real-time motion control, ROS (Robot Operating System), computer vision, and sophisticated sensor fusion algorithms to create a versatile robotic platform capable of autonomous navigation and environmental perception.

Key Innovation Areas

1. Autonomous Navigation: SLAM-based localization and simultaneous mapping
2. Real-time Control: 100+ Hz control loop with feedback compensation
3. Modular Architecture: Plug-and-play hardware and software modules
4. ROS Integration: Full Robot Operating System ecosystem compatibility
5. Computer Vision: Integration with RGB-D cameras and LIDAR sensors
6. Sensor Fusion: Multi-sensor fusion for robust localization and mapping

System Architecture

Hardware Platform

Primary Computing: BeagleBone Blue (AM5728 Dual-Core 1.5GHz)

• Onboard Features:
  • Dual H-bridge motor drivers
  • 8× servo PWM outputs
  • Multiple sensor interfaces
  • Built-in wireless (WiFi/Bluetooth)
  • Real-time processing capability

Secondary Compute: Raspberry Pi 3 (optional)

• Handles computationally intensive vision tasks
• ROS secondary node for distributed processing
• Enhanced storage and development flexibility

Mobility System

Drive Configuration: Differential drive with dual reduction gearmotors

Wheel Assembly

• Type: Omnidirectional wheels with caster balance wheel
• Diameter: 70mm (drive wheels), 25mm (caster)
• Traction: Rubber tread for indoor/outdoor operation
• Ground Clearance: 30mm (obstacle-capable)

Control Electronics

Motor Control Subsystem

Encoder Input (64 CPR, Quadrature)
    ↓
Speed Estimation (Low-Pass Filter)
    ↓
PID Controller (Kp=?, Ki=?, Kd=?)
    ↓
PWM Signal (16-bit, 1 kHz)
    ↓
H-Bridge Driver (Polulu DRV8835)
    ↓
Brushed DC Motor (6-12V)

Sensor Integration

Encoders: Quadrature shaft encoders for odometry

• Resolution: 64 counts per revolution
• Error Analysis: < 2% cumulative drift over 100m
• Update Rate: Continuous (event-driven)

Optional Sensors:

• LIDAR: 2D scanning for obstacle detection (URG-04LX compatible)
• RGB-D Camera: Depth-based object recognition
• IMU: Accelerometer/Gyroscope for tilt compensation
• Battery Monitor: Voltage/Current sensing for power management

Software Architecture

ROS Ecosystem Integration

Core Nodes:

1. motor_driver: Hardware abstraction for motor control
2. odometry: Position estimation from encoder feedback
3. navigation: Autonomous path planning and obstacle avoidance
4. vision: Image processing and object detection
5. state_machine: High-level behavior controller

Control Stack

Low-Level: Real-time control loop running on BeagleBone Blue

• Motor PWM commands (update rate: 100 Hz)
• Encoder reading and processing
• Safety watchdog monitoring

Mid-Level: ROS coordination layer

• Sensor fusion and filtering
• Trajectory computation
• Communication management

High-Level: Autonomous behavior framework

• Task planning and sequencing
• Decision-making algorithms
• Human-robot interaction

Motion Control Analysis

Speed Dynamics

The platform exhibits first-order speed response characteristics:

Where time constant τ ≈ 0.2 seconds, enabling rapid acceleration changes.

Trajectory Execution

Demonstrated capabilities:

• Square Pattern: 1m × 1m with < 5cm positioning error
• Curved Path: Smooth bezier curve following with adaptive speed
• Rotation: Full 360° in-place rotation with < 10° heading error
• Multi-segment: Complex paths combining translation and rotation

Performance Metrics

Autonomous Features

Navigation Algorithms

SLAM Implementation:

• Graph-based optimization (pose graph structure)
• Loop closure detection for map consistency
• Incremental mapping with bounded computation time

Path Planning:

• Dijkstra’s algorithm for global routes
• Dynamic Window Approach (DWA) for local collision avoidance
• Velocity obstacle (VO) computation for multi-robot scenarios

Computer Vision

Object Recognition:

• Trained YOLOv3/v4 models for common warehouse objects
• Transfer learning from COCO dataset
• Real-time inference on Raspberry Pi GPU acceleration

Pose Estimation:

• Marker-based AprilTag detection
• Feature-based visual odometry (ORB-SLAM compatible)
• Sensor fusion with encoder odometry

Experimental Results

Navigation Accuracy

Testing in structured indoor environment (5m × 5m):

• Systematic Error: < 3% of distance traveled
• Stochastic Error: < 2cm standard deviation
• Loop Closure: < 10cm position error after 50m circuit

Power Consumption

Battery capacity: 5 Ah @ 12V (60 Wh nominal)

Software Development Kit

ROS Packages and Architecture

The Hykabaï platform is built on standard ROS packages for navigation and perception:

Navigation Stack:

• move_base: Navigation framework for autonomous movement
• gmapping: SLAM implementation using Rao-Blackwellized particle filters
• amcl: Adaptive Monte Carlo localization for map-based navigation
• dwa_local_planner: Dynamic Window Approach for local collision avoidance

Vision Processing:

• image_transport: Efficient image streaming across ROS network
• opencv_ros: Computer vision pipeline integration
• depth_image_proc: Depth image processing from RGB-D sensors

Control and Monitoring:

• Custom motor control nodes running on BeagleBone Blue
• Battery monitoring and power management nodes
• Sensor aggregation and odometry computation

Key Integration Points

The platform integrates with standard ROS tools for development:

• ROS Master: Centralized communication hub for all nodes
• TF (Transform) Framework: Multi-dimensional transformation support
• rViz: Real-time visualization of sensor data and navigation state
• ROS Bag: Recording and playback of sensor data for offline analysis

Design Evolution

The Hykabaï platform represents the culmination of multiple design iterations:

v1-v2: Conceptual feasibility studies with simple differential drive v3-v5: Hardware integration and motor control refinement v6-v10: ROS ecosystem integration and basic autonomous navigation v11-v14: Computer vision pipeline development and SLAM implementation v14+: Advanced features including multi-task planning and machine learning

Current version v28 (CAD) and v3 (Physical) represent production-ready configurations.

Applications

Research Domains

• Autonomous robotics and navigation algorithms
• SLAM and sensor fusion techniques
• Computer vision and object detection
• Robot learning and adaptive control
• Human-robot interaction and social robotics
• Multi-robot coordination and swarms

Practical Applications

• Indoor autonomous navigation research
• Mobile platform sensor integration testing
• Autonomous mapping and exploration
• Sensor data collection and analysis
• Mobile robot algorithm development and validation

Future Enhancements

Planned Improvements:

1. Enhanced sensor suite (multimodal LIDAR, thermal imaging)
2. Improved SLAM algorithms with loop closure optimization
3. Advanced computer vision for dynamic object tracking
4. Deep learning-based end-to-end navigation
5. Reinforcement learning for task optimization
6. Extended battery life with hybrid power systems
7. ROS 2.0 migration for improved performance
8. Distributed computing with additional compute nodes
9. Advanced obstacle avoidance in crowded environments
10. Real-time performance monitoring and self-diagnostics

Technical Specifications Summary

Resources

• System Functional Diagram
• Complete CAD Model (3MF)
• Detailed CAD Assembly (F3Z)
• Electrical Schematics
• Technical Documentation Library