© 2026 Neuromorphic Humanoid Robotics

Researh Projects

Research Projects | NHURO Nexus

NHURO Research Sandbox

University of Aizu | Advanced Computer Systems Lab | AY2026 Graduation Thesis Framework

📂 Master SDK Source Hierarchy (Base Structure)

Students must adhere to this architecture. New nodes and interfaces should be integrated into the specified directories to ensure system compatibility.

nhuro_project/
├── voice_bridge.py             <-- (T1) Sensory Gateway (System Python)
└── nhuro_ws/                   <-- ROS 2 Workspace (Mamba Env)
    └── src/
        ├── nhuro_interfaces/   <-- ADD NEW .msg / .action FILES HERE
        │   ├── action/ Wave.action, Walk.action
        │   └── msg/
        └── nhuro_driver/       <-- ADD NEW RESEARCH NODES HERE
            └── nhuro_driver/
                ├── voice_parser_node.py   <-- (T2) The Brain
                ├── nhuro_action_node.py   <-- (T3) The Muscles
                └── nhuro/                 <-- Hardware Library
                    ├── robot.py
                    └── bus_servo.py

Detailed setup instructions are available in Module 0: Nexus Initialization.

🛠️ Available Laboratory Resources:
  • Sensors: Raspberry Pi Camera Module, USB Microphones, MPU6050 IMU.
  • Actuators: Bus-Servos (ttyAMA0), Motorized Active Grasper (Replacement Kit).
  • Software Environment: ROS 2 Humble, PyTorch, YOLOv8-tiny, Librosa.
AI & COMPUTER VISION

Task 1 (NHURO-Vision): Edge-AI Based Real-Time Object Recognition

This project addresses the limitation of hard-coded environmental sensing. Students must enable NHURO to identify and track dynamic objects in real-time using limited onboard compute.

Project Blueprint & New Files:
  • nhuro_interfaces/msg/Detection.msg: Define custom data types for object labels and bounding boxes.
  • nhuro_driver/vision_node.py: Implement a node that wraps the YOLOv8-tiny inference engine.
  • Design Hint: Optimize the model for the Raspberry Pi and bridge output to the nhuro_voice_text topic.
Evaluation Standards: Inference Speed (FPS), Mean Average Precision (mAP), and CPU utilization.
References for Task 1: 📄 You Only Look Once: Unified, Real-Time Object Detection 📄 YOLO: Real-Time Object Detection Framework 📄 Energy-Efficient Optimal Mode Selection for Edge AI Inference 📄 Design and Control of a Small Humanoid Equipped With Flight Unit and Wheels
ROBOTIC NAVIGATION & SPATIAL AI

Task 2 (NHURO-Nav): 3D Spatial Awareness & Autonomous Pathfinding

Building on Task 1, students must transition from 2D semantic detection to 3D spatial reasoning to enable NHURO to map its environment and navigate toward targets while avoiding obstacles.

Project Roadmap:
  • nhuro_driver/depth_node.py: Implement monocular depth estimation to transform 2D pixel data into 3D coordinates.
  • nhuro_driver/costmap_node.py: Project YOLO detection coordinates into a nav2_costmap_2d grid to define navigation zones.
  • Design Hint: Calculate the 3D centroid of detected objects to trigger locomotion goals rather than pre-programmed movement.
Evaluation Standards: Localization Precision (cm), Navigation Path Efficiency, and Collision-Free Success Rate.
References for Task 2: 📄 ROS 2 Navigation 2 (Nav2) Documentation 📄 FastDepth: Fast Monocular Depth Estimation 📄 Deep Learning for Visual Navigation in Humanoid Robotics 📄 Segment Anything for Autonomous Driving (Spatial-Semantic Integration)
CONTROL & KINEMATICS

NHURO-Dance: Development of a Rhythmic Motion Synchronization Framework for Multimodal Humanoid Interaction

Problem Description

The challenge is to align high-latency ROS 2 action execution with low-latency audio beats to create fluid, rhythmic entertainment routines.

Project Blueprint & New Files:
  • nhuro_driver/choreography_node.py: Create an Action Client that sequences multiple goals (Wave, Walk, Bow).
  • nhuro_driver/audio_analyzer.py: Implement real-time beat detection using librosa.
  • Design Hint: Adjust action step_duration parameters dynamically based on detected BPM.
Evaluation Standards: Rhythmic correctness (ms deviation from beat) and transition smoothness between discrete actions.
References: 📄 Key Pose-Based Dynamic Humanoid Motion Generation Using Parallel Multi-Fidelity Model Predictive Control (IEEE) 📄 ROS 2 Control: Synchronizing Multiple Actuators 📄 Autonomous Humanoid Robot Dance Generation System 📄 Dance Motion Control of a Humanoid Robot Based on Real-Time Tempo Tracking from Musical Audio Signals
ASSISTIVE MANIPULATION

NHURO-Care: Integration and Force-Limited Control of an Active Bionic Grasper for Assistive Humanoid Tasks

Problem Description

This project focuses on the integration of a motorized active grasper to allow for physical interaction with light objects like medicine bottles.

Project Blueprint & New Files:
  • nhuro_interfaces/action/Grasp.action: Define the goal (open/close) and feedback (pressure/torque).
  • nhuro_driver/nhuro/robot.py: Update the hardware library with a move_grasper() method.
  • Design Hint: Physically swap the static hand for the motorized replacement kit and map the new Servo ID.
Evaluation Standards: Grasp success rate (%), load-bearing capacity (grams), and torque accuracy to prevent crushing objects.
Academic References: 📄 Design of Low-Cost Bionic Grippers for Assistive Robots 📄 Force Control in Humanoid Hand Manipulation (IEEE) 📄 Multifingered force-aware control for humanoid robots 📄 Multifingered Robot Hand Compliant Manipulation Based on Human‑in‑the‑Loop Learning‑Control
NEURAL NETWORKS & CONTROL

NHURO-Brain: Design of an Artificial Neural Network Controller for Dynamic Gait Optimization in Bipedal Humanoids

Problem Description

Hard-coded patterns are unstable on uneven terrain. The student must develop a neural controller that adjusts the gait dynamically based on IMU feedback.

Project Blueprint & New Files:
  • nhuro_driver/imu_listener_node.py: Collect orientation data ($pitch$, $roll$) from the MPU6050.
  • nhuro_driver/neural_gait_node.py: Run a trained ANN to predict optimal servo angle offsets.
  • Design Hint: Train the model using PyTorch and deploy for real-time inference on the Pi.
Evaluation Standards: Balance stability (variance in IMU data), inference latency (ms), and successful walk distance without falling.
Academic References: 📄 Design of central pattern generator for humanoid robot walking based on multi-objective GA (IEEE) 📄 ZNN‑Based Gait Optimization for Humanoid Robots with ALIP and Inequality Constraints 📄 Reinforcement Learning for Versatile, Dynamic, and Robust Bipedal Locomotion Control 📄 Bipedal Robot Gait Control Strategies Based on Reinforcement Learning
HUMAN-ROBOT INTERACTION

NHURO-Interface: Development of a Low-Latency Web Dashboard and Digital Twin for Remote Humanoid Telemetry

Problem Description

Students must build a low-latency web dashboard to visualize “internal thoughts” (voice logs) and “physical state” (step count).

Project Blueprint & New Files:
  • ~/nhuro_project/dashboard/app.js: Implement a WebSocket client using roslibjs.
  • ~/nhuro_project/dashboard/index.html: Design a visual UI to show step counts and voice logs.
  • Design Hint: Use the rosbridge_suite to communicate between the browser and the ROS workspace.
Evaluation Standards: UI Refresh rate (Hz), data synchronization accuracy, and cross-compatibility.
Academic References: 📄 Introducing Robotics UI: A Web Interface Solution for ROS 2 Robots 📄 Documentation: Robot Web Tools (Humble) 📄 Low‑Latency Communications for Digital Twin Empowered Systems 📄 2. High‑Speed and Impact‑Resilient Teleoperation of Humanoid Robots
ROBOTIC DYNAMICS & SAFETY

NHURO-Safe: Autonomous Fall Detection and Recovery Strategies for Humanoid Robots via Inertial Sensory Feedback

Problem Description

Bipedal robots are prone to falling. The student must detect falls and execute autonomous “Stand-Up” sequences via IMU monitoring.

Project Blueprint & New Files:
  • nhuro_driver/fall_manager_node.py: Monitor pitch/roll thresholds to detect balance loss.
  • nhuro_interfaces/action/StandUp.action: Define the multi-phase stand-up sequence.
  • Design Hint: Develop high-torque servo patterns to safely return the robot to a neutral pose.
Evaluation Standards: Detection latency (ms), success rate of stand-up recovery, and prevention of servo overheating.
Academic References: 📄 Fall Detection and Standing Up Strategy for Humanoids 📄 Reflex Balance Schemes and Fast Fall Impact Mitigation Mechanisms 📄 Multi‑Phase Trajectory Optimization for Post‑Fall Autonomous Humanoid Recovery 📄 Whole‑Body Control for Humanoid Fall Mitigation
LLM & AUTONOMOUS REASONING

NHURO-LLM: LLM-Guided Autonomy for Neuromorphic Humanoid Robotics via Semantic Task Mapping

Problem Description

This project explores the integration of Large Language Models (LLMs) to bridge the gap between high-level natural language instructions and low-level, event-driven motor control. Students must develop a semantic translation layer that parses human intent into structured ROS 2 action goals.

Project Blueprint & New Files:
  • nhuro_driver/llm_bridge_node.py: Implement a node that interprets natural language, performs task decomposition, and publishes to action servers.
  • nhuro_interfaces/action/TaskExecution.action: Define an interface to handle complex task sequences (e.g., “Find the cup, then grasp it”).
  • Design Hint: Use structured output (JSON) from an LLM API and implement a state machine that validates safety parameters (e.g., pressure_limit) before dispatching commands.
Evaluation Standards: Semantic accuracy of task mapping, system latency (ms) from voice input to action, and strict adherence to hardware safety constraints during execution.
Academic References: 📄 Large Language Models for Robotics: Opportunities, Challenges, and Perspectives (arXiv) 📄 Vision-Language-Action Models for Robotics: A Review Towards Real-World Applications (IEEE) 📄 TinyVLA: Toward Fast, Data-Efficient Vision-Language-Action Models (IEEE RA-L) 📄 Exploring the Adversarial Vulnerabilities of Vision-Language-Action Models (arXiv)
← Return to NHURO Nexus ortal
© 2026 NHURO Nexus | Advanced Computing Systems Lab | Abderazek Ben Abdallah, Z. Wang

© 2026 Neuromorphic Humanoid Robotics. Created with ❤ using WordPress and Kubio