© 2018 by Dr. Alex Ramirez-Serrano

Humanoid Project

3D Printed life-size Humanoid project
High-speed control of humanoids for humanitarian aid.

In robot assisted Urban Search and Rescue (USAR) missions as well as in many other applications (e.g., mining), Autonomous Robotic Systems (AuRoS) are required to navigate and search disaster environments for victims, identify areas of concern, perform repair activities, etc. Since

these environments are often unknown and possess rough-terrain characteristics, it is decidedly challenging for these robots to carry

out their desired function. In this project, an architecture for a

system that enables AuRoS and humanoids to navigate in a

dynamic, rough terrain environment by adapting its locomotion as

needed (e.g. walking, crawling, and climbing) are being developed. The overall architecture will allow real-time processing of raw sensor data (e.g. position, orientation, joint position and velocities, and LiDAR depth maps) using deep reinforcement learning (DRL) to control the robot via a robot operating system (ROS) interface to improve its stability and agility for rough-terrain navigation. The implementation of this proposed system will rely on a number of existing DRL solutions including works related to rough terrain navigation using point-cloud mapping data, low-level feedback controllers for biped and other AuRoS locomotion, and example guided learning of physics-based character skills. The development of this system will involve iterating on both an accurate virtual model and the physical humanoid robot to validate its performance empirically.

Driven by the goal to develop tools to enhance trust and acceptance of AuRoS this project is developing a set of movement (e.g., via multi-point contact control) and intelligent tools to balance human and machine, opportunities and risks, automation and augmentation to build humanoid robots that Sense-Think-Act in real-time. The particular goals are to enhance the robot’s motion (locomotion, stability, speed), and sense mechanisms via a combination of machine deep learning, improved spatial cognition, and behavior libraries coupled with kinematic/dynamic modeling of compliant robotic systems. The specific goal is to enable AuRoS to be deployed in complex confined unstructured environments via effective navigation, high action speeds, and safe maneuvering / cooperation abilities. The project aims at a qualitative jump forward in robotic motor skills towards biological richness.

Biological walking systems (e.g., birds, dogs) can adaptively use their interlimb coordination for locomotion to deal with different situations such as flying through varying geometries small openings or crawling inside confined spaces. Achieving this on humanoids remains a challenge. We are developing revolutionary autonomous motion control, and cooperative strategies to meet human comparable efficiency.

This will be achieved using the robot's available resources (e.g., knees, arms, elbows) and kinematic/dynamic motion characteristics to maximize the robot’s agility/adaptability in all phases of its motion & interaction with humans.

Based on previous AuRoS developments conducted in our laboratory will are developing a set of robotic modules targeted to enable humanoids to autonomously move at high speeds within confined unstructured GPS-denied spaces where effective real-time sense and avoid as well as sophisticated locomotion tasks must be performed, a key for enhancing the capabilities of humanoids to be deployed without a predefined set of control instructions to be followed. 

The project is working on extending the motion characteristics of articulated robots (humanoids) and extending such abilities to include the use of the robot’s appendages and body parts to aid in the maneuvering phases.

The work will comprise experimental tests via the develpment of a state-of-the-art live-size humanoid robot. The mechanisms being developed are targeting complex confined spaces including moving at high speeds on moving surfaces (e.g., tilting platforms) and increasing the speed at which current humanoids are able to perform simple tasks, such as going up and down typical stairs and going through doors, and maneuvering through complex spaces (e.g., climbing ladders, or crawling through tight complex openings. 

The robot being developed within the context of this project employs a combination of Dynamixel MX and PRO motors (200, 100 and 20 Watt motors) which is enabling us to develop a life-size (170 cm tall) humanoid robot with 31 Degrees of Freedom (DoF), and having the capability to move in similar ways to humans and use human-centered tools (e.g. power tools, screwdrivers) if and when needed to accomplish the given mission (e.g., rescue a victim trapped inside a collapsed infrastructure).

Project Timeline:

The timeline below provides an overview of the developments that have taken place since 2016 in terms of the mechanical desing of a compliant highly-maneuverable life-size humanoid robot:

  • 2016:   First life-size 3D printed robot torso created:

    • using MX-28 Dynamixel motors

    • Manufactured with ABS 3D printed parts and liquid urethane

  • 2016:   v1.0 of 3D printed underactuated 3-finger robot hands developed

    • using MX-28 Dynamixel motors

    • controlled via tendons

  • 2017:   v1.0 of 6 DoF robot legs

    • using 200-watt Dynamixel PRO motors

    • retrofitted with ATI mini force/torque sensors

    • Manufactured with ABS 3D printed parts and Carbon fiber reinforcements

  • 2017:   Developed new 3-finger with 3 DoF robot hands

    • using RX Dynamixel motors

    • 3D printed in tough resing

  • 2018:   v2.0 of 7 DoF robot arms

    • using 100-watt Dynamixel PRO motors

    • Added ATI mini Torque/Force sensors on wrists

  • 2018:   Second generation of articulated robot feet

    • 3D printed with Onyx and carbon fiber parts

    • Added an ATI mini Torque/Force sensor

  • 2018:   v1.0 of 6 DoF robot legs

    • using 200-watt Dynamixel PRO motors

    • Manufactured with Aluminum, ABS, Onyx and carbon fiber 3D printed parts

  • 2018:   Development of a new robot head and neck mechanism

    • using 20-watt Dynamixel PRO motors

    • Manufactured with PLA 3D printed parts

    • Sensors used: Velodyne puck Lite Lidar, and INTEL T265 tracking camera

  • 2019:   Third generation cond robot ankles generated having 2 DoF

    • using 200 and 100-watt Dynamixel PRO motors

    • ankles include compliant mechanisms

  • 2019:   2nd version of robot hip with 5 DoF

    • using 200 and 100-watt Dynamixel PRO motors

    • Manufactured using Aluminum plates and 3D printed parts

2018

Developed articulated feet with passive articulation

2016

Robot Torso using 100% 3D printed parts

2019

New hip and ankles 

2017

Added 3D printed legs using Onyx, ABS and Carbon Fiber

Robots are coming to the rescue!