Computational Intelligence applied to Motion Planning and Control in Biped Robotics: A survey Juan J. Figueredo

Structure 1. Introduction „ 2. Dynamical and Biological Aspects of BipedWalking „ 3. Problem Definition „ 4. A taxonomy of motion planning and control methods in biped walking „ 5. Discussion and future trends „

Introduction „

Motion planning and control is relevant because: … Most

environments are devised to adapt well to humans. … It gives us understanding of human morphology, mechanics and control. … It is a complex control problem in nonlinear and nonholonomic systems.

Biomechanical Preliminaries of Walking „

Modeling Walking: … Articulated

motion of several rigid bodies … Simplification to 12-DOF in lower limbs. „

„ „

„

In the hip: Flexion-Extension, Abduction-Adduction, ExternalInternal Rotation In the knee: Flexion-Extension In the ankle: Plantarflexion-Dorsiflexion, PronationSupination

Stability … Static … Dynamic

Stability

CoM

ZMP

Biological Considerations on Biped Locomotion „

Six models of bipedal walking (Vaughan): … Bipedal

walking as an evolutionary adaptation of hominids. … Minimization of energy consumption by displacing the CoM along an optimal path … Progressive learning with risk of falling minimization … Spinal cord interneurons acting as rhythmic central pattern generators … Neural system training along with biomechanical system and environment adaptation … Feedback control in powered dynamic locomotion

Biological Considerations on Biped Locomotion (II) „ „ „

„

Tree levels for neural motion control: Feedback control in motoneurons: contribution of reflex action over motoneuron signal intensity. Feedback control in central pattern generator flexor-extensor centers: movement synchronization and reflex response to perturbation and loads. Higher level control: conscious control of locomotion..

Problem Definition „

Give to a biped robot an optimal locomotion … Conserve

static stability … Walk conserving dynamic stability … Locally minimize energy consumption in locomotion … React to external perturbations … Plan gait trajectories to attain specific objectives … Move across unstructured environments … Globally minimize energy consumption in locomotion for a given objective

A taxonomy of motion planning and control methods in biped walking

Central Pattern Generator (CPG) Methods „

„

„

Coordinated rhythmic stimulation to generate a gait pattern. It is inspired from the spinal motor center find in animals. Approaches: …

Oscillatory Neural Networks … Evolutionary Optimization of CPGs … Rhythmical Dynamic Systems

Trajectory Tracking Methods Generating a kinematic pattern in a way such that following it (in joint spaced) yields a successful gait pattern. „ Advances in energy consumption optimization. „ Approaches: „

… Predefined

trajectories … Evolved Trajectories

Dynamic Walking Control „ „ „ „ „

Obtain dynamic stable walking by a control based on various techniques. Highly complex. ZMP related. Partial objective tracking and phase reset techniques. Approaches: … Recurrent

Neural Networks … Fuzzy Systems … Hybrid Methods

Static Walking Control „ „ „

„

Displaced by Dynamic Control Very low velocities required to make negligible the inertial forces. They still have a research value in studying the stationary phases and posture of bipeds and some specific movement patterns. Approaches: … Neural

Networks … Genetic Algorithms

Discussion and future trends „ „ „

„

„

Current development in the field of biped robotics is highly varied Several objectives cannot be completely fulfilled be one method alone Active control … Remarkable achievements … High inefficiency … Cannot perform well in unstructured environments Passive Dynamic Walking … High efficiencies but cannot navigate satisfactorily in 2D and 3D Computational Intelligence … In early development … Flexibility … Adaptation … Low energy consumption