DEXIGRAB

**/ˈdɛk.si.ɡræb/ (DEK-see-grab) - DEXIGRABs (plural): /ˈdɛk.si.ɡræbz/ (DEK-see-grabz)

An advanced, libre-source robot hand with integrated hybrid soft robotics, and soon GelSense-touch capabilities with pretrained sensor fusion RL models

Introduction

"What? Another esoteric 3D printed robot hand project that's basically just a toy/puppet/useless gizmo dohicky with 5 string motors? No! There's too many of those already!"

Well, here's where I might change your mind regarding that!

Features

• Flexible TPU palm mechanics, allowing for hybrid soft robotics that allow for actual grasping around curved objects in comparison to current traditional robotic palm mechanics
• 16 motors allowing for 16 DOF hand dexerity, 3 DOF for each finger and 4 DOF on the thumb
• Straightforward 3D printing and assembly with only 5-10 3D printed parts to be assembled together for more plug-and-play use dynamics
• Mechanics and electronics entirely constrained inside the palm, allowing for full end-effector modularity with all current and future robotic manipulators
• To utilize a modified L3-F-TOUCH sensing module for the fingers and palm and a stereo raspberry pi camera system for sensor fusion data to train reliable deep reinforcement learning
• Universal Robot Attachment system that can be 3D printed to interface with any robot arm's end effector attachment system
• Multi-modal sensor fusion integrating:
  • Intel® RealSense™ Depth Camera D455
  • L3-F-TOUCH sensing for tactile feedback
  • Arduino Pro Mini for motor encoder data collection

AI & Simulation

• Proximal Policy Optimization (PPO) Deep Reinforcement Learning algorithm for training
• NVIDIA's Isaac Labs and Isaac Sim for physics-accurate simulation and training, with Genesis Physics Engine examples in the near future (batch training on hybrid robots still a work in-progress, see lines 179-180 since Genesis 0.2.1: https://github.com/Genesis-Embodied-AI/Genesis/blob/main/genesis/engine/simulator.py)
• Upcoming trained AI/RL models to be released allowing for automated grasping capabilities of generalized objects

Related Resources

• L3-F-TOUCH shown
• General Concept of touch module
• Simulation environment to be possibly integrated

Applications

• Primary use as a robotic end effector with advanced grasping capabilities
• Secondary application as a low-weight and cost-accessible hand for prosthetics design, especially if built with worm drive gearboxes for human-level gripping strength
• Customizable socket system (hopefully also full copyleft) for any given person

Project Details

• <$500 cost to build
• Fully open source hardware and software under the GPL 3.0 and CERN-OHL pair license :)

Coming Soon

• Assembly instructions
• Component lists
• Software setup guide
• Simulation integration details
• Attachment system designs