Merge pull request #4182 from AndrewMeadows/isentropic

fix bad integration of rapidly spinning objects
This commit is contained in:
Brad Hefta-Gaub 2015-01-27 18:33:31 -08:00
commit 8994b9f51c
7 changed files with 134 additions and 32 deletions

View file

@ -14,6 +14,7 @@
#include <ByteCountCoding.h>
#include <GLMHelpers.h>
#include <Octree.h>
#include <PhysicsHelpers.h>
#include <RegisteredMetaTypes.h>
#include <SharedUtil.h> // usecTimestampNow()
@ -711,10 +712,22 @@ void EntityItem::simulateKinematicMotion(float timeElapsed) {
} else {
// NOTE: angularSpeed is currently in degrees/sec!!!
// TODO: Andrew to convert to radians/sec
float angle = timeElapsed * glm::radians(angularSpeed);
glm::vec3 axis = _angularVelocity / angularSpeed;
glm::quat dQ = glm::angleAxis(angle, axis);
glm::quat rotation = glm::normalize(dQ * getRotation());
glm::vec3 angularVelocity = glm::radians(_angularVelocity);
// for improved agreement with the way Bullet integrates rotations we use an approximation
// and break the integration into bullet-sized substeps
glm::quat rotation = getRotation();
float dt = timeElapsed;
while (dt > PHYSICS_ENGINE_FIXED_SUBSTEP) {
glm::quat dQ = computeBulletRotationStep(angularVelocity, PHYSICS_ENGINE_FIXED_SUBSTEP);
rotation = glm::normalize(dQ * rotation);
dt -= PHYSICS_ENGINE_FIXED_SUBSTEP;
}
// NOTE: this final partial substep can drift away from a real Bullet simulation however
// it only becomes significant for rapidly rotating objects
// (e.g. around PI/4 radians per substep, or 7.5 rotations/sec at 60 substeps/sec).
glm::quat dQ = computeBulletRotationStep(angularVelocity, dt);
rotation = glm::normalize(dQ * rotation);
setRotation(rotation);
}
}

View file

@ -15,6 +15,7 @@
#include "BulletUtil.h"
#include "EntityMotionState.h"
#include "PhysicsEngine.h"
#include "PhysicsHelpers.h"
QSet<EntityItem*>* _outgoingEntityList;

View file

@ -14,6 +14,7 @@
#include "BulletUtil.h"
#include "ObjectMotionState.h"
#include "PhysicsEngine.h"
#include "PhysicsHelpers.h"
const float DEFAULT_FRICTION = 0.5f;
const float MAX_FRICTION = 10.0f;
@ -107,6 +108,7 @@ bool ObjectMotionState::shouldSendUpdate(uint32_t simulationFrame) {
if (_sentFrame == 0) {
_sentPosition = bulletToGLM(_body->getWorldTransform().getOrigin());
_sentVelocity = bulletToGLM(_body->getLinearVelocity());
_sentRotation = bulletToGLM(_body->getWorldTransform().getRotation());
_sentAngularVelocity = bulletToGLM(_body->getAngularVelocity());
_sentFrame = simulationFrame;
return false;
@ -118,7 +120,8 @@ bool ObjectMotionState::shouldSendUpdate(uint32_t simulationFrame) {
glm::vec3 wasAngularVelocity = _sentAngularVelocity;
#endif
float dt = (float)(simulationFrame - _sentFrame) * PHYSICS_ENGINE_FIXED_SUBSTEP;
int numFrames = simulationFrame - _sentFrame;
float dt = (float)(numFrames) * PHYSICS_ENGINE_FIXED_SUBSTEP;
_sentFrame = simulationFrame;
bool isActive = _body->isActive();
@ -170,16 +173,16 @@ bool ObjectMotionState::shouldSendUpdate(uint32_t simulationFrame) {
if (glm::length2(_sentAngularVelocity) > 0.0f) {
// compute rotation error
_sentAngularVelocity *= powf(1.0f - _angularDamping, dt);
float spin = glm::length(_sentAngularVelocity);
const float MIN_SPIN = 1.0e-4f;
if (spin > MIN_SPIN) {
glm::vec3 axis = _sentAngularVelocity / spin;
_sentRotation = glm::normalize(glm::angleAxis(dt * spin, axis) * _sentRotation);
float attenuation = powf(1.0f - _angularDamping, dt);
_sentAngularVelocity *= attenuation;
// Bullet caps the effective rotation velocity inside its rotation integration step, therefore
// we must integrate with the same algorithm and timestep in order achieve similar results.
for (int i = 0; i < numFrames; ++i) {
_sentRotation = glm::normalize(computeBulletRotationStep(_sentAngularVelocity, PHYSICS_ENGINE_FIXED_SUBSTEP) * _sentRotation);
}
}
const float MIN_ROTATION_DOT = 0.98f;
const float MIN_ROTATION_DOT = 0.99f; // 0.99 dot threshold coresponds to about 16 degrees of slop
glm::quat actualRotation = bulletToGLM(worldTrans.getRotation());
#ifdef WANT_DEBUG

View file

@ -11,6 +11,7 @@
#include "PhysicsEngine.h"
#include "ShapeInfoUtil.h"
#include "PhysicsHelpers.h"
#include "ThreadSafeDynamicsWorld.h"
static uint32_t _numSubsteps;
@ -29,23 +30,27 @@ PhysicsEngine::~PhysicsEngine() {
// begin EntitySimulation overrides
void PhysicsEngine::updateEntitiesInternal(const quint64& now) {
// NOTE: the grand order of operations is:
// (1) relay incoming changes
// (2) step simulation
// (3) synchronize outgoing motion states
// (4) send outgoing packets
// this is step (4)
QSet<ObjectMotionState*>::iterator stateItr = _outgoingPackets.begin();
while (stateItr != _outgoingPackets.end()) {
ObjectMotionState* state = *stateItr;
if (state->doesNotNeedToSendUpdate()) {
stateItr = _outgoingPackets.erase(stateItr);
} else if (state->shouldSendUpdate(_numSubsteps)) {
state->sendUpdate(_entityPacketSender, _numSubsteps);
++stateItr;
} else {
++stateItr;
// no need to send updates unless the physics simulation has actually stepped
if (_lastNumSubstepsAtUpdateInternal != _numSubsteps) {
_lastNumSubstepsAtUpdateInternal = _numSubsteps;
// NOTE: the grand order of operations is:
// (1) relay incoming changes
// (2) step simulation
// (3) synchronize outgoing motion states
// (4) send outgoing packets
// this is step (4)
QSet<ObjectMotionState*>::iterator stateItr = _outgoingPackets.begin();
while (stateItr != _outgoingPackets.end()) {
ObjectMotionState* state = *stateItr;
if (state->doesNotNeedToSendUpdate()) {
stateItr = _outgoingPackets.erase(stateItr);
} else if (state->shouldSendUpdate(_numSubsteps)) {
state->sendUpdate(_entityPacketSender, _numSubsteps);
++stateItr;
} else {
++stateItr;
}
}
}
}

View file

@ -14,8 +14,6 @@
#include <stdint.h>
const float PHYSICS_ENGINE_FIXED_SUBSTEP = 1.0f / 60.0f;
#include <QSet>
#include <btBulletDynamicsCommon.h>
@ -114,6 +112,7 @@ private:
ContactMap _contactMap;
uint32_t _numContactFrames = 0;
uint32_t _lastNumSubstepsAtUpdateInternal = 0;
};
#endif // hifi_PhysicsEngine_h

View file

@ -0,0 +1,58 @@
//
// PhysicsHelpers.cpp
// libraries/shared/src
//
// Created by Andrew Meadows 2015.01.27
// Unless otherwise copyrighted: Copyright 2015 High Fidelity, Inc.
//
// Unless otherwise licensced: Distributed under the Apache License, Version 2.0.
// See the accompanying file LICENSE or http://www.apache.org/licenses/LICENSE-2.0.html
//
#include "PhysicsHelpers.h"
#include "SharedUtil.h"
// This chunk of code was copied from Bullet-2.82, so we include the Bullet license here:
/*
* Bullet Continuous Collision Detection and Physics Library
* Copyright (c) 2003-2009 Erwin Coumans http://bulletphysics.org
*
* This software is provided 'as-is', without any express or implied warranty.
* In no event will the authors be held liable for any damages arising from the use of this software.
* Permission is granted to anyone to use this software for any purpose,
* including commercial applications, and to alter it and redistribute it freely,
* subject to the following restrictions:
*
* 1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software.
* If you use this software in a product, an acknowledgment in the product documentation would be appreciated but
* is not required.
* 2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
* 3. This notice may not be removed or altered from any source distribution.
*
* Copied and modified from LinearMath/btTransformUtil.h by AndrewMeadows on 2015.01.27.
* */
glm::quat computeBulletRotationStep(const glm::vec3& angularVelocity, float timeStep) {
// Bullet uses an exponential map approximation technique to integrate rotation.
// The reason for this is to make it easy to compute the derivative of angular motion for various consraints.
// Here we reproduce the same approximation so that our extrapolations agree well with Bullet's integration.
//
// Exponential map
// google for "Practical Parameterization of Rotations Using the Exponential Map", F. Sebastian Grassia
float speed = glm::length(angularVelocity);
// limit the angular motion because the exponential approximation fails for large steps
const float ANGULAR_MOTION_THRESHOLD = 0.5f * PI_OVER_TWO;
if (speed * timeStep > ANGULAR_MOTION_THRESHOLD) {
speed = ANGULAR_MOTION_THRESHOLD / timeStep;
}
glm::vec3 axis = angularVelocity;
if (speed < 0.001f) {
// use Taylor's expansions of sync function
axis *= (0.5f * timeStep - (timeStep * timeStep * timeStep) * (0.020833333333f * speed * speed));
} else {
// sync(speed) = sin(c * speed)/t
axis *= (sinf(0.5f * speed * timeStep) / speed );
}
return glm::quat(cosf(0.5f * speed * timeStep), axis.x, axis.y, axis.z);
}

View file

@ -0,0 +1,23 @@
//
// PhysicsHelpers.h
// libraries/shared/src
//
// Created by Andrew Meadows 2015.01.27
// Copyright 2015 High Fidelity, Inc.
//
// Distributed under the Apache License, Version 2.0.
// See the accompanying file LICENSE or http://www.apache.org/licenses/LICENSE-2.0.html
//
#ifndef hifi_PhysicsHelpers_h
#define hifi_PhysicsHelpers_h
#include <glm/glm.hpp>
#include <glm/gtc/quaternion.hpp>
const float PHYSICS_ENGINE_FIXED_SUBSTEP = 1.0f / 60.0f;
// return incremental rotation (Bullet-style) caused by angularVelocity over timeStep
glm::quat computeBulletRotationStep(const glm::vec3& angularVelocity, float timeStep);
#endif // hifi_PhysicsHelpers_h