ephysics/ephysics/engine/ConstraintSolver.hpp

139 lines
7.1 KiB
C++

/** @file
* @author Daniel Chappuis
* @copyright 2010-2016 Daniel Chappuis
* @license BSD 3 clauses (see license file)
*/
#pragma once
#include <ephysics/configuration.hpp>
#include <ephysics/mathematics/mathematics.hpp>
#include <ephysics/constraint/Joint.hpp>
#include <ephysics/engine/Island.hpp>
#include <etk/Map.hpp>
namespace ephysics {
/**
* This structure contains data from the constraint solver that are used to solve
* each joint constraint.
*/
struct ConstraintSolverData {
public :
float timeStep; //!< Current time step of the simulation
vec3* linearVelocities; //!< Array with the bodies linear velocities
vec3* angularVelocities; //!< Array with the bodies angular velocities
vec3* positions; //!< Reference to the bodies positions
etk::Quaternion* orientations; //!< Reference to the bodies orientations
const etk::Map<RigidBody*, uint32_t>& mapBodyToConstrainedVelocityIndex; //!< Reference to the map that associates rigid body to their index in the constrained velocities array
bool isWarmStartingActive; //!< True if warm starting of the solver is active
/// Constructor
ConstraintSolverData(const etk::Map<RigidBody*, uint32_t>& refMapBodyToConstrainedVelocityIndex):
linearVelocities(nullptr),
angularVelocities(nullptr),
positions(nullptr),
orientations(nullptr),
mapBodyToConstrainedVelocityIndex(refMapBodyToConstrainedVelocityIndex) {
}
};
/**
* @brief This class represents the constraint solver that is used to solve constraints between
* the rigid bodies. The constraint solver is based on the "Sequential Impulse" technique
* described by Erin Catto in his GDC slides (http://code.google.com/p/box2d/downloads/list).
*
* A constraint between two bodies is represented by a function C(x) which is equal to zero
* when the constraint is satisfied. The condition C(x)=0 describes a valid position and the
* condition dC(x)/dt=0 describes a valid velocity. We have dC(x)/dt = Jv + b = 0 where J is
* the Jacobian matrix of the constraint, v is a vector that contains the velocity of both
* bodies and b is the constraint bias. We are looking for a force F_c that will act on the
* bodies to keep the constraint satisfied. Note that from the virtual work principle, we have
* F_c = J^t * lambda where J^t is the transpose of the Jacobian matrix and lambda is a
* Lagrange multiplier. Therefore, finding the force F_c is equivalent to finding the Lagrange
* multiplier lambda.
* An impulse P = F * dt where F is a force and dt is the timestep. We can apply impulses a
* body to change its velocity. The idea of the Sequential Impulse technique is to apply
* impulses to bodies of each constraints in order to keep the constraint satisfied.
*
* --- Step 1 ---
*
* First, we int32_tegrate the applied force F_a acting of each rigid body (like gravity, ...) and
* we obtain some new velocities v2' that tends to violate the constraints.
*
* v2' = v1 + dt * M^-1 * F_a
*
* where M is a matrix that contains mass and inertia tensor information.
*
* --- Step 2 ---
*
* During the second step, we iterate over all the constraints for a certain number of
* iterations and for each constraint we compute the impulse to apply to the bodies needed
* so that the new velocity of the bodies satisfy Jv + b = 0. From the Newton law, we know that
* M * deltaV = P_c where M is the mass of the body, deltaV is the difference of velocity and
* P_c is the constraint impulse to apply to the body. Therefore, we have
* v2 = v2' + M^-1 * P_c. For each constraint, we can compute the Lagrange multiplier lambda
* using : lambda = -m_c (Jv2' + b) where m_c = 1 / (J * M^-1 * J^t). Now that we have the
* Lagrange multiplier lambda, we can compute the impulse P_c = J^t * lambda * dt to apply to
* the bodies to satisfy the constraint.
*
* --- Step 3 ---
*
* In the third step, we int32_tegrate the new position x2 of the bodies using the new velocities
* v2 computed in the second step with : x2 = x1 + dt * v2.
*
* Note that in the following code (as it is also explained in the slides from Erin Catto),
* the value lambda is not only the lagrange multiplier but is the multiplication of the
* Lagrange multiplier with the timestep dt. Therefore, in the following code, when we use
* lambda, we mean (lambda * dt).
*
* We are using the accumulated impulse technique that is also described in the slides from
* Erin Catto.
*
* We are also using warm starting. The idea is to warm start the solver at the beginning of
* each step by applying the last impulstes for the constraints that we already existing at the
* previous step. This allows the iterative solver to converge faster towards the solution.
*
* For contact constraints, we are also using split impulses so that the position correction
* that uses Baumgarte stabilization does not change the momentum of the bodies.
*
* There are two ways to apply the friction constraints. Either the friction constraints are
* applied at each contact point or they are applied only at the center of the contact manifold
* between two bodies. If we solve the friction constraints at each contact point, we need
* two constraints (two tangential friction directions) and if we solve the friction
* constraints at the center of the contact manifold, we need two constraints for tangential
* friction but also another twist friction constraint to prevent spin of the body around the
* contact manifold center.
*/
class ConstraintSolver {
private :
const etk::Map<RigidBody*, uint32_t>& m_mapBodyToConstrainedVelocityIndex; //!< Reference to the map that associates rigid body to their index in the constrained velocities array
float m_timeStep; //!< Current time step
bool m_isWarmStartingActive; //!< True if the warm starting of the solver is active
ConstraintSolverData m_constraintSolverData; //!< Constraint solver data used to initialize and solve the constraints
public :
/// Constructor
ConstraintSolver(const etk::Map<RigidBody*, uint32_t>& _mapBodyToVelocityIndex);
/// Initialize the constraint solver for a given island
void initializeForIsland(float _dt, Island* _island);
/// Solve the constraints
void solveVelocityConstraints(Island* _island);
/// Solve the position constraints
void solvePositionConstraints(Island* _island);
/// Return true if the Non-Linear-Gauss-Seidel position correction technique is active
bool getIsNonLinearGaussSeidelPositionCorrectionActive() const {
return m_isWarmStartingActive;
}
/// Enable/Disable the Non-Linear-Gauss-Seidel position correction technique.
void setIsNonLinearGaussSeidelPositionCorrectionActive(bool _isActive) {
m_isWarmStartingActive = _isActive;
}
/// Set the constrained velocities arrays
void setConstrainedVelocitiesArrays(vec3* _constrainedLinearVelocities,
vec3* _constrainedAngularVelocities);
/// Set the constrained positions/orientations arrays
void setConstrainedPositionsArrays(vec3* _constrainedPositions,
etk::Quaternion* _constrainedOrientations);
};
}