Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Empowering Humanoid Robots for Stable Balancing

Dissertation Proposal

Author: Roberto Figueiredo

Supervisor: Prof. Doutor Nuno Lau

Co-Supervisor: Prof. Doctor Daniel Polani

Masters in Robotics and Intelligent Systems

Empowering Humanoid Robots for Stable Balancing | Thesis proposal
  1. Introduction
    • Evolution from WABOT-1 to Atlas
    • The progression of humanoid robotics
  2. Problem Definition
    • Challenges in dynamic environments and limitations of traditional control methods
  3. Motivation
    • The importance of intrinsic motivations and the application of biological concepts in robotics
  4. Theoretical Framework
    • Distinction between intrinsic and extrinsic motivations
    • Introduction to the concept of empowerment
  5. Empowerment in Humanoid Robotics
    • Definition and calculation of empowerment
    • Application of empowerment for stable balancing
  6. State of the Art
    • Review of current methodologies: Linear Inverted Pendulum Model, Model Predictive Control, Whole Body Control
  7. Proposed Solution
    • Outline of the empowerment framework and its implementation
  8. Work Planning
    • Steps for developing and testing the empowerment framework
  9. Conclusion
    • Recap and invitation for questions
WABOT-1
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Introduction

Wabot-1

  • First anthropomorphic robot
  • Limb control | Vision | Conversation
Atlas
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Introduction

Atlas

  • Most advanced humanoid robot
  • Advanced perception and control
  • Complex motion
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

It is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility

Hans Moravex

Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Problem

  • Dynamic Environments
  • Unpredictable and constantly changing
  • Complex and unstable structure of humanoid robots
  • Large number of DoF, high center of gravity
  • Limitations of traditional control methods
  • Need for extensive calibration, tunning and training
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Motivation

  • Exploring Intrinsic Motivations in Humanoid Robotics
    To investigate how intrinsic motivations can be applied and enhance the learning capabilities of humanoid robots.

  • Biological Concept Application
    Expand the research and exploration of Biological concepts in AI and robotics.

  • RoboCup as a Test Bed
    Utilizing the RoboCup competition to test and demonstrate the potential of stable balancing in humanoid robots through the application of intrinsic motivations.

Motivations
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Theoretical framework

Intrinsic vs Extrinsic motivations

  • Extrinsic Motivations
    • External rewards
    • Money, grades, praise, etc.
  • Intrinsic Motivations
    • Internal rewards
    • Enjoyment, curiosity, etc.
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Special case of Empowerment

  • Measure of the potential for an agent to influence its environment
  • Intrinsic motivation
  • Encourages the agent to explore and learn about its environment
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Empowerment definition

Empowerment is calculated as the mutual information between the agent's actions and the resultant states, given the current state

Quantifies how much knowing one variable informs us about another.

Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Empowerment definition

Using Gaussian channels, the empowerment can be calculated as the channel capacity.

also, the empowerment can be calculated as the maximum rate at which information can be transmitted over a channel.

Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Empowerment Calculation

  1. Linear Response Approximation
  2. Singular Value Decomposition
  3. Waterfilling Algorithm
  4. Empowerment Calculation

Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Objectives and Contributions

  • Review and analyze current technologies and methodologies for humanoid robot balancing
  • Develop a framework for the application of empowerment motivations in humanoid robots
  • Simulate and Evaluate the performance of the proposed framework
  • Empirical testing of the framework in a humanoid robot
  • Contribute to the body of knowledge in the field of humanoid robotics and AI
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

State of the Art

Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Linear Inverted Pendulum Model

  • Common model for humanoid robot balancing

  • Uses the pendulum model to represent the robot's center of mass.

  • Simplified model that assumes the robot's center of mass moves in a straight line

  • Often used as a basis for more complex control algorithms.

  • EROS team, used feet pressure sensors to estimate the CoP

Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Linear Inverted Pendulum Model

Results

First Image
Second Image

Comparison of the CoP estimation with and without feet pressure sensors

Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Model Predictive Control

CP-MPC Framework, A Model Predictive Capture Point Control Framework. Via Ankle, Hip and Stepping strategies.

  • Variable weighting methods for CAM
  • Objective: Enhance balancing by fine-tuning the CAM influence on the control output.
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Model Predictive Control

Framework

Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Model Predictive Control

Three strategies

  • ZMP Control and Stepping Control
    Handling the Zero Moment Point (ZMP) control and stepping control.
  • Variable weighting methods for CAM
    Adjusts the weighting parameters of CAM damping control during the MPC time horizon to enhance CP control
  • Hierarchical control structure of MPC
    Enables optimization of step time
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Whole Body Control

Ballancing is formulated hierarchically.
Each layer is formulated as a QP problem.
The hierarchy ensures the prioritization of higher level tasks.

  1. Floating Base Dynamics Handles the overall dynamics of the robot's floating base, essential for maintaining balance during movement and interaction with the environment.
  2. Foot Position and Posture Manages the positioning and orientation of the feet to ensure stable contact with the ground, adjusting as necessary for different terrains or motions.
  3. Linear Momentum Controls the robot's linear momentum to prevent falls by adjusting its center of mass and motion trajectory.
  4. Torso Posture and Foot Contact Force Regulates the upper body posture and distributes forces across the feet to maintain stability.
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Whole Body Control

  • Requires high speed computation
  • High bandwidth communication
  • 1ms to resolve 4 QP to maintain a 1kHz control loop
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Tools

Pinocchio

Open Source C++ library for fast and efficient rigid body dynamics computations and their analytical derivatives.

Enables precise modeling, simulation and control of humanoid robots.

Efficient, as it can generate robot-specific dynamic models at compile time.

Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Proposed solution

Empowerment Framework

  • Application of Empowerment to Humanoid Balancing
  • Utilizing the Pinocchio library for the dynamic model
  • Simulate and Evaluate the performance of the proposed framework
Empowering Humanoid Robots for Stable Balancing | Thesis proposal

Work Plan

  • Setup Virtual Environment
  • Test baseline control methods
  • Implement Pinocchio
  • Research and Implement Empowerment Framework
  • Simulate and Evaluate
  • Empirical Testing
  • Dissertation Writing

    • Empowering Humanoid Robots for Stable Balancing | Thesis proposal

    Thank You

    Questions?

    update design

    **Challenges in Humanoid Balancing** - **Motivation for Research** - **Contribution and Objectives**

    Humanoid robotics had a first breakthrough in 1970 with the development of WABOT-1

    It was able to perform basic motion and walking

    - **Challenges in Humanoid Balancing** - **Motivation for Research** - **Contribution and Objectives**

    We now have very advanced robots such as Atlas, probably the most famous

    Capable of doing complex perception tasks and motions such as backflips

    Shannon-Hartley theorem, used for a single channel

    maxp(x) indicates that we're maximizing over all possible probability distributions p(x) of the random variable X

    When calculating empowerment, the goal is to find the particular action distribution p(x) that, when the agent follows it, leads to the maximum mutual information between the actions taken and the subsequent states of the system. This means that the chosen actions should lead to outcomes that are as predictable as possible, given the state of the system when the actions are taken. In a more empowered state, an agent's actions lead to specific, predictable outcomes, which indicates that the agent has a high degree of control over its environment. The action distribution that maximizes this control is the one that maximizes the mutual information, indicating a strong and clear relationship between what the agent does and what happens as a result.

    The Capture Point (CP) is a point on the ground where a humanoid robot must place its foot in order to come to a complete stop without falling. It's calculated based on the robot's current momentum and its dynamic model.

    Cendtroidal Angular Momentum (CAM) rotational momentum of a system about its center of mass

    ZMP - It specifies the point with respect to which reaction forces at the contacts between the feet and the ground do not produce any moment in the horizontal direction

    ZMP - Zero moment point - hankle

    CAM - HIP

    Footstep - feet

    400hz by bitbots using dynamixel

    - Use RoboCup as a test bed for empirical testing