Omnidirectional Humanoid Robot Balance Control via Linear Predictive MPC
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current humanoid robots lack the ability to autonomously adapt their behavior to dynamic environments without programmer intervention, struggling to maintain balance and stability when disturbances occur, particularly when tilting, due to limitations in controlling the Center of Pressure (CoP) and managing discrete contact forces.
Innovation Solution
A humanoid robot with an omnidirectional mobile base, equipped with sensors and actuators, employs a dual-state control system using a no-tilt and tilt-state controller, along with a landing-state controller, based on a double point-mass robot model and linear model predictive control, to manage CoP and angular velocity, ensuring stability and smooth transitions between states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a standard PID controller is used to maintain robot balance, then the control implementation is simple, but the robot cannot effectively handle disturbances and tilting situations
Solution Approach 1:
The patent transitions from a static PID controller to a dynamic Model Predictive Controller that adapts its control strategy based on the robot's current state. The controller dynamically switches between no-tilt mode and tilt mode depending on whether the CoP is within or outside the support polygon, enabling effective handling of disturbances while maintaining computational tractability through state-based dynamic adaptation.
Solution Approach 2:
The patent changes the control parameters dynamically by switching between different control models based on the robot's tilt state. When the CoP is within the support polygon, a standard MPC model is used; when the CoP exits the support polygon, a tilt-specific MPC model with adjusted parameters is activated, allowing the system to maintain reliability under varying disturbance conditions.
2Stability of the object's composition
If the CoP is constrained to remain strictly inside the support polygon, then the robot maintains dynamic stability, but the robot loses the ability to recover from disturbances that push the CoP to the boundary
Solution Approach 1:
The patent prepares for potential disturbances by maintaining the CoP within the support polygon during normal operation, ensuring stability. When a disturbance occurs and the CoP approaches or exits the support polygon boundary, the system activates a recovery mode that generates corrective torques to bring the CoP back inside, thus preliminarily preparing the system to handle boundary conditions while maintaining overall stability.
Solution Approach 2:
The patent converts the potentially harmful condition of CoP exiting the support polygon (which would normally cause falling) into a beneficial recovery opportunity. By detecting when the CoP is outside the support polygon and activating a specific tilt recovery controller, the system uses this boundary condition as a trigger for corrective action, transforming the harmful disturbance into a controlled recovery process that enhances the robot's adaptability.
3Reliability
If a linear predictive controller is used to predict future robot states, then the robot can proactively maintain stability, but the computational complexity increases
Solution Approach 1:
The patent segments the control problem into distinct operational modes (no-tilt mode and tilt mode) based on the CoP position relative to the support polygon. This segmentation allows the use of simplified predictive models for each mode rather than a single complex universal model, reducing overall computational complexity while maintaining proactive stability maintenance through state-based model selection.
Solution Approach 2:
The patent applies partial predictive control by using a linear predictive model only for the dominant stability maintenance task, rather than implementing a full non-linear predictive controller. This partial action approach provides sufficient proactive stability for the robot's operational requirements while keeping computational complexity manageable through the use of linear approximations and state-based simplifications.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The object of the invention is a humanoid robot (100) with a body (190) joined to an omnidirectional mobile ground base (140), equipped with : - a body position sensor, a base position sensor and an angular velocity sensor to provide measures, - actuators (212) comprising at least 3 wheels located in the omnidirectional mobile base, - extractors (211) for converting sensored measures into useful data, - a supervisor (500) to calculate position, velocity and acceleration commands from the useful data, - means for converting commands into instructions for the actuators, characterized in that the supervisor comprises: - a no-tilt state controller (501), a tilt state controller (502) and a landing state controller (503), each controller comprising means for calculating, position, velocity and acceleration commands based on a double point-mass robot model with tilt motion and on a linear model predictive control law, expressed as a quadratic optimization formulation with a weighted sum of objectives, and a set of predefined linear constraints.