MIT 2.120 – Introduction to Robotics (CV-based Manipulator Gripper)

Winner of Spring 2022 2.12/2.120 Robotic Challenge Competition!

Team receiving “Best Project” award at MIT 2.12/2.120 Robotic Challenge Competition

Project Requirements

Utilizing a Universal Robots UR5e Collaborative Robot…

  • Identify and pick up large scale construction bricks from a pile of randomly placed bricks
  • Orient and position bricks over a worksite
  • Place bricks sequentially to build up a perimeter wall around the worksite

Hardware Design

Cross-sectional view of custom manipulator gripper design

Key Design Features

  • Computer Vision (CV) achieved through embedded camera ESP32 micro-controller package
  • Redundant axial plane HX-711 load cell for contact sensing
  • D-shaped gripper with compliant sleeves for multi-optional gripping postures
  • Radial gear rack and pinion assembly powered by a servo for gripper actuation
  • Modular plate-by-plate mechanical assembly approach to enable easy component replacement/troubleshooting
Cross-sectional view of Fused Filament Fabrication (FFF) print plan

Design considerations

Servo motor provides single point-of-failure for system

  • Ran lifecycle tests on gripper actuation and validated approximate lifetime of actuator to be 250 cycles (2.5 times higher than required lifecycle)
Lifecycle testing of gear rack and pinion actuation

Load cell provides possible false readings or noise that can interrupt the operation

  • High-pass filter integrated into data-processing node for load cell, tests conclude that readings fall within acceptable threshold for contact sensing
  • High-level control architecture designed to accept load cell readings upon failed CV depth sensing

Software Design

Software architecture for robotic gripper system

Key Design Features

  • Developed in ROS/rosserial package, enabling cross-compatibility framework with UR5e ROS driver
  • Utilized overhead Intel Realsense camera for inertial tracking/redundant depth sensing
  • Object detection achieved through HSV threshold
  • Simultaneous state tracking and trajectory planning via ROS node with UR5e
  • Object positioning and orientation discovery achieved through edge tracking with bounding box

Design considerations

ESP32 camera reliability is questionable

  • Utilize overhead Realsense camera as redundant visualizer and extend object detection and location identification algorithm

Location of placed bricks must be tracked to avoid unintentional stacking order

  • Before executing brick placement step, utilize Realsense camera to check location of placed bricks and reconfigure placement location upon discrepancy discovery

Brick falling from gripper during transit

  • Run “IS_GRABBED” sequence to check for successful lift and hold of brick via Realsense and ESP32 cameras