rtabmap/corelib/src/OptimizerGTSAM.cpp

/*
Copyright (c) 2010-2014, Mathieu Labbe - IntRoLab - Universite de Sherbrooke
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
    * Redistributions of source code must retain the above copyright
      notice, this list of conditions and the following disclaimer.
    * Redistributions in binary form must reproduce the above copyright
      notice, this list of conditions and the following disclaimer in the
      documentation and/or other materials provided with the distribution.
    * Neither the name of the Universite de Sherbrooke nor the
      names of its contributors may be used to endorse or promote products
      derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY
DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "rtabmap/core/Graph.h"

#include <rtabmap/utilite/ULogger.h>
#include <rtabmap/utilite/UStl.h>
#include <rtabmap/utilite/UMath.h>
#include <rtabmap/utilite/UConversion.h>
#include <rtabmap/utilite/UTimer.h>
#include <set>

#include <rtabmap/core/OptimizerGTSAM.h>

#ifdef WITH_GTSAM
#include <gtsam/geometry/Pose2.h>
#include <gtsam/geometry/Pose3.h>
#include <gtsam/inference/Key.h>
#include <gtsam/inference/Symbol.h>
#include <gtsam/slam/PriorFactor.h>
#include <gtsam/slam/BetweenFactor.h>
#include <gtsam/nonlinear/NonlinearFactorGraph.h>
#include <gtsam/nonlinear/GaussNewtonOptimizer.h>
#include <gtsam/nonlinear/DoglegOptimizer.h>
#include <gtsam/nonlinear/LevenbergMarquardtOptimizer.h>
#include <gtsam/nonlinear/NonlinearOptimizer.h>
#include <gtsam/nonlinear/Marginals.h>
#include <gtsam/nonlinear/Values.h>

#include "vertigo/gtsam/betweenFactorMaxMix.h"
#include "vertigo/gtsam/betweenFactorSwitchable.h"
#include "vertigo/gtsam/switchVariableLinear.h"
#include "vertigo/gtsam/switchVariableSigmoid.h"
#endif // end WITH_GTSAM

namespace rtabmap {

bool OptimizerGTSAM::available()
{
#ifdef WITH_GTSAM
	return true;
#else
	return false;
#endif
}

std::map<int, Transform> OptimizerGTSAM::optimize(
		int rootId,
		const std::map<int, Transform> & poses,
		const std::multimap<int, Link> & edgeConstraints,
		std::list<std::map<int, Transform> > * intermediateGraphes,
		double * finalError,
		int * iterationsDone)
{
	std::map<int, Transform> optimizedPoses;
#ifdef WITH_GTSAM
	UDEBUG("Optimizing graph...");
	if(edgeConstraints.size()>=1 && poses.size()>=2 && iterations() > 0)
	{
		gtsam::NonlinearFactorGraph graph;

		//prior first pose
		UASSERT(uContains(poses, rootId));
		const Transform & initialPose = poses.at(rootId);
		if(isSlam2d())
		{
			gtsam::noiseModel::Diagonal::shared_ptr priorNoise = gtsam::noiseModel::Diagonal::Sigmas(gtsam::Vector3(0.01, 0.01, 0.01));
			graph.add(gtsam::PriorFactor<gtsam::Pose2>(rootId, gtsam::Pose2(initialPose.x(), initialPose.y(), initialPose.theta()), priorNoise));
		}
		else
		{
			gtsam::noiseModel::Diagonal::shared_ptr priorNoise = gtsam::noiseModel::Diagonal::Sigmas((gtsam::Vector(6) << 1e-6, 1e-6, 1e-6, 1e-4, 1e-4, 1e-4).finished());
			graph.add(gtsam::PriorFactor<gtsam::Pose3>(rootId, gtsam::Pose3(initialPose.toEigen4d()), priorNoise));
		}

		UDEBUG("fill poses to gtsam...");
		gtsam::Values initialEstimate;
		for(std::map<int, Transform>::const_iterator iter = poses.begin(); iter!=poses.end(); ++iter)
		{
			UASSERT(!iter->second.isNull());
			if(isSlam2d())
			{
				initialEstimate.insert(iter->first, gtsam::Pose2(iter->second.x(), iter->second.y(), iter->second.theta()));
			}
			else
			{
				initialEstimate.insert(iter->first, gtsam::Pose3(iter->second.toEigen4d()));
			}
		}

		UDEBUG("fill edges to gtsam...");
		int switchCounter = poses.rbegin()->first+1;
		for(std::multimap<int, Link>::const_iterator iter=edgeConstraints.begin(); iter!=edgeConstraints.end(); ++iter)
		{
			int id1 = iter->first;
			int id2 = iter->second.to();

			UASSERT(!iter->second.transform().isNull());

			if(this->isRobust() &&
			   iter->second.type()!=Link::kNeighbor &&
			   iter->second.type() != Link::kNeighborMerged)
			{
				// create new switch variable
				// Sunderhauf IROS 2012:
				// "Since it is reasonable to initially accept all loop closure constraints,
				//  a proper and convenient initial value for all switch variables would be
				//  sij = 1 when using the linear switch function"
				double prior = 1.0;
				initialEstimate.insert(gtsam::Symbol('s',switchCounter), vertigo::SwitchVariableLinear(prior));

				// create switch prior factor
				// "If the front-end is not able to assign sound individual values
				//  for Ξij , it is save to set all Ξij = 1, since this value is close
				//  to the individual optimal choice of Ξij for a large range of
				//  outliers."
				gtsam::noiseModel::Diagonal::shared_ptr switchPriorModel = gtsam::noiseModel::Diagonal::Sigmas(gtsam::Vector1(1.0));
				graph.add(gtsam::PriorFactor<vertigo::SwitchVariableLinear> (gtsam::Symbol('s',switchCounter), vertigo::SwitchVariableLinear(prior), switchPriorModel));
			}

			if(isSlam2d())
			{
				Eigen::Matrix<double, 3, 3> information = Eigen::Matrix<double, 3, 3>::Identity();
				if(!isCovarianceIgnored())
				{
					// For some reasons, dividing by 1000 avoids some exceptions (maybe too large numbers on optimization)
					information(0,0) = iter->second.infMatrix().at<double>(0,0)/1000.0; // x-x
					information(0,1) = iter->second.infMatrix().at<double>(0,1)/1000.0; // x-y
					information(0,2) = iter->second.infMatrix().at<double>(0,5)/1000.0; // x-theta
					information(1,0) = iter->second.infMatrix().at<double>(1,0)/1000.0; // y-x
					information(1,1) = iter->second.infMatrix().at<double>(1,1)/1000.0; // y-y
					information(1,2) = iter->second.infMatrix().at<double>(1,5)/1000.0; // y-theta
					information(2,0) = iter->second.infMatrix().at<double>(5,0)/1000.0; // theta-x
					information(2,1) = iter->second.infMatrix().at<double>(5,1)/1000.0; // theta-y
					information(2,2) = iter->second.infMatrix().at<double>(5,5)/1000.0; // theta-theta
				}
				gtsam::noiseModel::Gaussian::shared_ptr model = gtsam::noiseModel::Gaussian::Information(information);

				if(this->isRobust() &&
				   iter->second.type()!=Link::kNeighbor &&
				   iter->second.type() != Link::kNeighborMerged)
				{
					// create switchable edge factor
					graph.add(vertigo::BetweenFactorSwitchableLinear<gtsam::Pose2>(id1, id2, gtsam::Symbol('s', switchCounter++), gtsam::Pose2(iter->second.transform().x(), iter->second.transform().y(), iter->second.transform().theta()), model));
				}
				else
				{
					graph.add(gtsam::BetweenFactor<gtsam::Pose2>(id1, id2, gtsam::Pose2(iter->second.transform().x(), iter->second.transform().y(), iter->second.transform().theta()), model));
				}
			}
			else
			{
				Eigen::Matrix<double, 6, 6> information = Eigen::Matrix<double, 6, 6>::Identity();
				if(!isCovarianceIgnored())
				{
					memcpy(information.data(), iter->second.infMatrix().data, iter->second.infMatrix().total()*sizeof(double));
					// For some reasons, dividing by 1000 avoids some exceptions (maybe too large numbers on optimization)
					information = information / 1000.0;
				}

				gtsam::noiseModel::Gaussian::shared_ptr model = gtsam::noiseModel::Gaussian::Information(information);

				if(this->isRobust() &&
				   iter->second.type()!=Link::kNeighbor &&
				   iter->second.type() != Link::kNeighborMerged)
				{
					// create switchable edge factor
					graph.add(vertigo::BetweenFactorSwitchableLinear<gtsam::Pose3>(id1, id2, gtsam::Symbol('s', switchCounter++), gtsam::Pose3(iter->second.transform().toEigen4d()), model));
				}
				else
				{
					graph.add(gtsam::BetweenFactor<gtsam::Pose3>(id1, id2, gtsam::Pose3(iter->second.transform().toEigen4d()), model));
				}
			}
		}

		UDEBUG("create optimizer");
		gtsam::GaussNewtonParams parameters;
		parameters.relativeErrorTol = epsilon();
		parameters.maxIterations = iterations();
		gtsam::GaussNewtonOptimizer optimizer(graph, initialEstimate, parameters);
		//gtsam::LevenbergMarquardtParams parametersLev;
		//parametersLev.relativeErrorTol = epsilon();
		//parametersLev.maxIterations = iterations();
		//gtsam::LevenbergMarquardtOptimizer optimizer(graph, initialEstimate, parametersLev);
		//gtsam::DoglegParams parametersDogleg;
		//parametersDogleg.relativeErrorTol = epsilon();
		//parametersDogleg.maxIterations = iterations();
		//gtsam::DoglegOptimizer optimizer(graph, initialEstimate, parametersDogleg);

		UINFO("GTSAM optimizing begin (max iterations=%d, robust=%d)", iterations(), isRobust()?1:0);
		UTimer timer;
		int it = 0;
		double lastError = 0.0;
		for(int i=0; i<iterations(); ++i)
		{
			if(intermediateGraphes && i > 0)
			{
				std::map<int, Transform> tmpPoses;
				for(gtsam::Values::const_iterator iter=optimizer.values().begin(); iter!=optimizer.values().end(); ++iter)
				{
					if(iter->value.dim() > 1)
					{
						if(isSlam2d())
						{
							gtsam::Pose2 p = iter->value.cast<gtsam::Pose2>();
							tmpPoses.insert(std::make_pair((int)iter->key, Transform(p.x(), p.y(), p.theta())));
						}
						else
						{
							gtsam::Pose3 p = iter->value.cast<gtsam::Pose3>();
							tmpPoses.insert(std::make_pair((int)iter->key, Transform::fromEigen4d(p.matrix())));
						}
					}
				}
				intermediateGraphes->push_back(tmpPoses);
			}
			try
			{
				optimizer.iterate();
				++it;
			}
			catch(gtsam::IndeterminantLinearSystemException & e)
			{
				UERROR("GTSAM exception catched: %s", e.what());
				return optimizedPoses;
			}

			// early stop condition
			double error = optimizer.error();
			UDEBUG("iteration %d error =%f", i+1, error);
			double errorDelta = lastError - error;
			if(i>0 && errorDelta < this->epsilon())
			{
				if(errorDelta < 0)
				{
					UDEBUG("Negative improvement?! Ignore and continue optimizing... (%f < %f)", errorDelta, this->epsilon());
				}
				else
				{
					UINFO("Stop optimizing, not enough improvement (%f < %f)", errorDelta, this->epsilon());
					break;
				}
			}
			else if(i==0 && error < this->epsilon())
			{
				UINFO("Stop optimizing, error is already under epsilon (%f < %f)", error, this->epsilon());
				break;
			}
			lastError = error;
		}
		if(finalError)
		{
			*finalError = lastError;
		}
		if(iterationsDone)
		{
			*iterationsDone = it;
		}
		UINFO("GTSAM optimizing end (%d iterations done, error=%f (initial=%f final=%f), time=%f s)", optimizer.iterations(), optimizer.error(), graph.error(initialEstimate), graph.error(optimizer.values()), timer.ticks());

		for(gtsam::Values::const_iterator iter=optimizer.values().begin(); iter!=optimizer.values().end(); ++iter)
		{
			if(iter->value.dim() > 1)
			{
				if(isSlam2d())
				{
					gtsam::Pose2 p = iter->value.cast<gtsam::Pose2>();
					optimizedPoses.insert(std::make_pair((int)iter->key, Transform(p.x(), p.y(), p.theta())));
				}
				else
				{
					gtsam::Pose3 p = iter->value.cast<gtsam::Pose3>();
					optimizedPoses.insert(std::make_pair((int)iter->key, Transform::fromEigen4d(p.matrix())));
				}
			}
		}
	}
	else if(poses.size() == 1 || iterations() <= 0)
	{
		optimizedPoses = poses;
	}
	else
	{
		UWARN("This method should be called at least with 1 pose!");
	}
	UDEBUG("Optimizing graph...end!");
#else
	UERROR("Not built with GTSAM support!");
#endif
	return optimizedPoses;
}

} /* namespace rtabmap */