Nirav Madhani
<- Back to Projects
Mar 1, 2023

Mobile Robotic Arm (ROS + DL Models)

RoboticsROSML/DLHardware

Challenges Solved

  • Perception-to-Actuation: Integrated Deep Learning perception models with ROS nodes for physical actuation.
  • Kinematics Synchronization: Solved real-world kinematics synchronization for a wheeled robotic arm platform.

Signal

Physical Robotics / Hardware Integration

Project Architecture

The system is built on a modular ROS architecture that decouples high-level perception from low-level motor control:

  • Perception Node: Runs real-time Deep Learning models (likely CNN-based) to identify target objects and calculate 3D coordinates.
  • Kinematics Node: Receives target coordinates and computes the necessary joint angles using a custom Inverse Kinematics solver.
  • Actuation Node: Communicates with hardware controllers (servos/motors) to execute the movements.

Technical Depth

  • Real-time Inference: Optimized models for low-latency inference to meet the control requirements of physical hardware.
  • Coupled Motion: Engineered a synchronization layer that accounts for the movement of the wheeled base, allowing the arm to track targets dynamically.
  • Hardware Integration: Directly interfaced with motor drivers and sensors via ROS serial and custom drivers.

Media

Links