Phillip Grote, Joaquim Ortiz-Haro, Marc Toussaint, and Ozgur S. Oguz.
Under review, 2024.
|
@misc{grote2023neural,
title = {Neural Field Representations of Articulated Objects for Robotic Manipulation Planning},
author = {Grote, Phillip and Ortiz-Haro, Joaquim and Toussaint, Marc and Oguz, Ozgur S.},
booktitle = {Under review},
year = {2024},
pdf = {https://arxiv.org/abs/2309.07620}
}
Khaled Wahba, Joaquim Ortiz-Haro, Marc Toussaint, and Wolfgang Hönig.
Under review, 2024.
|
@misc{wahba2023kinodynamic,
title = {Kinodynamic Motion Planning for a Team of Multirotors Transporting a Cable-Suspended Payload in Cluttered Environments},
author = {Wahba, Khaled and Ortiz-Haro, Joaquim and Toussaint, Marc and Hönig, Wolfgang},
booktitle = {Under review},
year = {2024},
pdf = {https://arxiv.org/abs/2310.03394}
}
Joaquim Ortiz-Haro, Wolfgang Hönig, Valentin N. Hartmann, Marc Toussaint, and Ludovic Righetti.
2024.
| VIDEO
| WEBPAGE
|
@misc{ortizharo2024idbrrt,
title = {iDb-RRT: Sampling-based Kinodynamic Motion Planning with Motion Primitives and Trajectory Optimization},
author = {Ortiz-Haro, Joaquim and Hönig, Wolfgang and Hartmann, Valentin N. and Toussaint, Marc and Righetti, Ludovic},
year = {2024},
eprint = {2403.10745},
archiveprefix = {arXiv},
primaryclass = {cs.RO},
pdf = {https://arxiv.org/abs/2403.10745},
webpage = {https://quimortiz.github.io/idbrrt},
youtube = {3ToQU-qLWg0}
}
Modern robots excel at performing simple and repetitive tasks in controlled environments; however, future applications, such as robotic construction and assistance, will require long-term planning of physical interactions.
These problems can be formulated as Task and Motion Planning (TAMP).
The goal is to find how the robot should move to solve complex tasks requiring multiple interactions with objects in the environment, such as building furniture or cleaning and organizing the kitchen. However, TAMP is notoriously difficult to solve because it involves a tight combination of task planning and motion planning, considering geometric and physical constraints.
In this thesis, we aim to improve the performance of TAMP algorithms from three complementary perspectives.
First, we investigate the integration of discrete task planning with continuous trajectory optimization.
Our main contribution is a conflict-based solver that automatically discovers why a task plan might fail when considering the constraints of the physical world.
This information is then fed back into the task planner, resulting in an efficient, bidirectional, and intuitive interface between task and motion, capable of solving TAMP problems with multiple objects, robots, and tight physical constraints.
Traditionally, there have been two competing approaches to solving TAMP problems: sample-based and optimization-based methods.
In the second part, we first illustrate that, given the wide range of tasks and environments within TAMP, neither sampling nor optimization is superior in all settings.
To combine the strengths of both approaches, we have designed meta-solvers for TAMP, adaptive solvers that automatically select which algorithms and computations to use and how to best decompose each problem to find a solution faster.
A third promising direction to improve TAMP algorithms is to learn from previous solutions to similar problems.
In the third part, we combine deep learning architectures with model-based reasoning to accelerate computations within our TAMP solver.
Specifically, we target infeasibility detection and nonlinear optimization, focusing on generalization, accuracy, compute time, and data efficiency.
At the core of our contributions is a refined, factored representation of the trajectory optimization problems inside TAMP.
This structure not only facilitates more efficient planning, encoding of geometric infeasibility, and meta-reasoning but also provides better generalization in neural architectures.
@phdthesis{ortiz2024factored,
title = {Factored Task and Motion Planning with Combined Optimization, Sampling and Learning},
author = {Ortiz-Haro, Joaquim},
pdf = {/assets/phd_thesis.pdf},
slides = {/assets/phd_slides.pdf},
booktitle = {TU-Berlin},
year = {2024}
}
Publications
Solving Sequential Manipulation Puzzles by Finding Easier Subproblems
Levit Svetlana, Joaquim Ortiz-Haro, and Marc Toussaint.
ICRA, 2024.
|
@misc{levit2023solving,
title = {Solving Sequential Manipulation Puzzles by Finding Easier Subproblems},
author = {Svetlana, Levit and Ortiz-Haro, Joaquim and Toussaint, Marc},
booktitle = {ICRA},
year = {2024}
}
Effort Level Search in Infinite Completion Trees with Application to Task-and-Motion Planning
Marc Toussaint, Joaquim Ortiz-Haro, Valentin Hartmann, Erez Karpas, and Wolfgang Hoenig.
ICRA, 2024.
|
@misc{toussaint2023effort,
title = {Effort Level Search in Infinite Completion Trees with Application to Task-and-Motion Planning},
author = {Toussaint, Marc and Ortiz-Haro, Joaquim and Hartmann, Valentin and Karpas, Erez and Hoenig, Wolfgang},
booktitle = {ICRA},
year = {2024}
}
Akmaral Moldagalieva, Joaquim Ortiz-Haro, Marc Toussaint, and Wolfgang Hönig.
ICRA, 2024.
|
@misc{moldagalieva2023dbcbs,
title = {db-CBS: Discontinuity-Bounded Conflict-Based Search for Multi-Robot Kinodynamic Motion Planning},
author = {Moldagalieva, Akmaral and Ortiz-Haro, Joaquim and Toussaint, Marc and Hönig, Wolfgang},
pdf = {https://arxiv.org/abs/2309.16445},
booktitle = {ICRA},
year = {2024}
}
Valentin N. Hartmann, Joaquim Ortiz-Haro, and Marc Toussaint.
IEEE Int. Conf. on Intelligent Robots and Systems (IROS), 2023.
|
@misc{https://doi.org/10.48550/arxiv.2303.00637,
doi = {10.48550/ARXIV.2303.00637},
url = {https://arxiv.org/abs/2303.00637},
pdf = {https://arxiv.org/abs/2303.00637},
author = {Hartmann, Valentin N. and Ortiz-Haro, Joaquim and Toussaint, Marc},
keywords = {Robotics (cs.RO), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {Efficient Path Planning In Manipulation Planning Problems by Actively Reusing Validation Effort},
year = {2023},
booktitle = {IEEE Int. Conf. on Intelligent Robots and Systems (IROS)},
copyright = {arXiv.org perpetual, non-exclusive license}
}
J. Ortiz-Haro, J.-S. Ha, D. Driess, E. Karpas, and M. Toussaint.
IEEE Int. Conf. on Robotics and Automation (ICRA), 2023.
|
@inproceedings{ortiz2022conflict,
author = {Ortiz-Haro, J. and Ha, J.-S. and Driess, D. and Karpas, E. and Toussaint, M.},
year = {2023},
title = {Learning Feasibility of Factored Nonlinear Programs in Robotic Manipulation Planning},
pdf = {https://arxiv.org/abs/2210.12386},
booktitle = {IEEE Int. Conf. on Robotics and Automation (ICRA)}
}
J. Ortiz-Haro, E. Karpas, M. Katz, and M. Toussaint.
IEEE Robotics and Automation Letters (RA-L), 2022.
| WEBPAGE
|
@inproceedings{ortiz2022conflictdriven,
author = {Ortiz-Haro, J. and Karpas, E. and Katz, M. and Toussaint, M.},
year = {2022},
pdf = {https://arxiv.org/abs/2211.15275},
title = {Conflict-driven Interface between Symbolic Planning and Nonlinear Constraint Solving},
booktitle = {IEEE Robotics and Automation Letters (RA-L)},
webpage = {https://quimortiz.github.io/graphnlp/}
}
Wolfgang Hönig, Joaquim Ortiz-Haro, and Marc Toussaint.
Proc. of the IEEE Int. Conf. on Intelligent Robots and Systems
(IROS), 2022.
|
@inproceedings{honig2022db,
title = {db-A*: Discontinuity-bounded Search for Kinodynamic Mobile Robot Motion Planning},
author = {Hönig, Wolfgang and Ortiz-Haro, Joaquim and Toussaint, Marc},
year = {2022},
pdf = {https://arxiv.org/abs/2203.11108},
booktitle = {Proc{.} of the IEEE Int{.} Conf{.} on Intelligent Robots and Systems
(IROS)}
}
Cornelius V. Braun, Joaquim Ortiz-Haro, Marc Toussaint, and Ozgur S. Oguz.
Proc. of the IEEE Int. Conf. on Intelligent Robots and Systems
(IROS), 2022.
|
|
Sequential decision-making and motion planning for robotic manipulation induce combinatorial complexity. For long-horizon tasks, especially when the environment comprises many objects that can be interacted with, planning efficiency becomes even more important. To plan such long-horizon tasks, we present the RHH-LGP algorithm for combined task and motion planning (TAMP). First, we propose a TAMP approach (based on Logic-Geometric Programming) that effectively uses geometry-based heuristics for solving long-horizon manipulation tasks. We further improve the efficiency of this planner by a receding horizon formulation, resulting in RHH-LGP. We demonstrate the effectiveness and generality of our approach on several long-horizon tasks that require reasoning about interactions with a large number of objects. Using our framework, we can solve tasks that require multiple robots, including a mobile robot and snake-like walking robots, to form novel heterogeneous kinematic structures autonomously.
@inproceedings{braun2022rhhlgp,
title = {RHH-LGP: Receding Horizon And Heuristics-Based Logic-Geometric Programming For Task And Motion Planning},
author = {Braun, Cornelius V. and Ortiz-Haro, Joaquim and Toussaint, Marc and Oguz, Ozgur S.},
year = {2022},
eprint = {2110.03420},
archiveprefix = {arXiv},
primaryclass = {cs.RO},
pdf = {https://arxiv.org/abs/2110.03420},
booktitle = {Proc{.} of the IEEE Int{.} Conf{.} on Intelligent Robots and Systems
(IROS)}
}
Jay Kamat, Joaquim Ortiz-Haro, Marc Toussaint, Florian T. Pokorny, and Andreas Orthey.
Proc. of the IEEE Int. Conf. on Intelligent Robots and Systems (IROS), 2022.
| WEBPAGE
|
@inproceedings{kamat2022bitkomo,
title = {BITKOMO: Combining Sampling and Optimization for Fast Convergence in Optimal Motion Planning},
author = {Kamat, Jay and Ortiz-Haro, Joaquim and Toussaint, Marc and Pokorny, Florian T. and Orthey, Andreas},
year = {2022},
pdf = {https://arxiv.org/abs/2203.01751},
booktitle = {Proc{.} of the IEEE Int{.} Conf{.} on Intelligent Robots and Systems (IROS)},
webpage = {https://sites.google.com/goa.bits-pilani.ac.in/bitkomo}
}
Joaquim Ortiz-Haro, Erez Karpas, Michael Katz, and Marc Toussaint.
International Conference on Automated Planning and Scheduling (ICAPS), 2022.
| WEBPAGE
|
|
Robots operating in the real world must combine task planning for reasoning about what to do with motion planning for reasoning about how to do it – this is known as task and motion planning. One promising approach for task and motion planning is Logic Geometric Programming (LGP) which integrates a logical layer and a geometric layer in an optimization formulation. The logical layer describes feasible high-level actions at an abstract symbolic level, while the geometric layer uses continuous optimization methods to reason about motion trajectories with geometric constraints.
In this paper we propose a new approach for solving task and motion planning problems in the LGP formulation, that leverages state-of-the-art diverse planning at the logical layer to explore the space of feasible logical plans, and minimizes the number of optimization problems to be solved on the continuous geometric layer.
To this end, geometric infeasibility is fed back into planning by identifying prefix conflicts and incorporating this back into the planner through a novel multi-prefix forbidding compilation. We further leverage diverse planning with a new novelty criteria for selecting candidate plans based on the prefix novelty, and a metareasoning approach which attempts to extract only useful conflicts by leveraging the information that is gathered in the course of solving the given problem.
@inproceedings{ortiz2022conflictdirected,
title = {Conflict-Directed Diverse Planning for Logic-Geometric Programming},
author = {Ortiz-Haro, Joaquim and Karpas, Erez and Katz, Michael and Toussaint, Marc},
booktitle = {International Conference on Automated Planning and Scheduling (ICAPS)},
year = {2022},
webpage = {/ConflictPlanningLGP/},
pdf = {https://ojs.aaai.org/index.php/ICAPS/article/view/19811}
}
Joaquim Ortiz-Haro, Jung-Su Ha, Danny Driess, and Marc Toussaint.
5th Annual Conference on Robot Learning (CoRL), 2021.
| POSTER
|
|
Sampling efficiently on constraint manifolds is a core problem in robotics.
We propose Deep Generative Constraint Sampling (DGCS), which combines a deep generative model for sampling close to a constraint manifold with nonlinear constrained optimization to project to the constraint manifold.
The generative model is conditioned on the problem instance, taking a scene image as input, and it is trained with a dataset of solutions and a novel analytic constraint term.
To further improve the precision and diversity of samples, we extend the approach to exploit a factorization of the constrained problem. We evaluate our approach in two problems of robotic sequential manipulation in cluttered environments.
Experimental results demonstrate that our deep generative model produces diverse and precise samples and outperforms heuristic warmstart initialization.
@inproceedings{ortiz2021structured,
title = {Structured deep generative models for sampling on constraint manifolds in sequential manipulation},
author = {Ortiz-Haro, Joaquim and Ha, Jung-Su and Driess, Danny and Toussaint, Marc},
booktitle = {5th Annual Conference on Robot Learning (CoRL)},
year = {2021},
pdf = {https://openreview.net/pdf?id=CPbn4N3a2zC},
poster = {/assets/poster_corl21.pdf}
}
Joaquim Ortiz-Haro, Valentin N. Hartmann, Ozgur S. Oguz, and Marc Toussaint.
Proc. of the IEEE Int. Conf. on Robotics and
Automation (ICRA), 2021.
| VIDEO
|
|
Efficient sampling from constraint manifolds, and thereby generating a diverse set of solutions for feasibility problems, is a fundamental challenge. We consider the case where a problem is factored, that is, the underlying nonlinear program is decomposed into differentiable equality and inequality constraints, each of which depends only on some variables. Such problems are at the core of efficient and robust sequential robot manipulation planning. Naive sequential conditional sampling of individual variables, as well as fully joint sampling of all variables at once (e.g., leveraging optimization methods), can be highly inefficient and non-robust. We propose a novel framework to learn how to break the overall problem into smaller sequential sampling problems. Specifically, we leverage Monte-Carlo Tree Search to learn assignment orders for the variable-subsets, in order to minimize the computation time to generate feasible full samples. This strategy allows us to efficiently compute a set of diverse valid robot configurations for mode-switches within sequential manipulation tasks, which are waypoints for subsequent trajectory optimization or sampling-based motion planning algorithms. We show that the learning method quickly converges to the best sampling strategy for a given problem, and outperforms user-defined orderings or fully joint optimization, while providing a higher sample diversity.
@inproceedings{21-ortiz-ICRA,
title = {Learning Efficient Constraint Graph Sampling for Robotic
Sequential Manipulation},
author = {Ortiz-Haro, Joaquim and Hartmann, Valentin N. and Oguz, Ozgur S. and Toussaint, Marc},
pdf = {https://arxiv.org/abs/2011.04828},
booktitle = {Proc{.} of the IEEE Int{.} Conf{.} on Robotics and
Automation (ICRA)},
year = {2021},
youtube = {mCNdvjTbHNI}
}
Joaquim Ortiz-Haro.
Research Report. Supervisor: Gergely Neu. University Pompeu Fabra (UPF), Barcelona, 2019.
|
|
The optimal transport distance is a powerful and useful metric to compare probability distributions in statistics and machine learning. In contrast to alternative distance measures, the main drawback is the computational complexity.
Adding entropic regularization to the original linear program, we can find approximate solutions much faster (reducing complexity from O(n3) to O(n2), where n is the histogram size) using the Sinkhorn algorithm (iterative matrix scaling).
However, finding solutions with small error requires using small regularizations, which increases the number of Sinkhorn iterations. The dependence is k = O(e-2), where k is the number of Sinkhorn iterations and e the approximate error.
In this report, we have proposed an algorithm based on proximal regularization. It consists on solving a sequence of problems decreasing the regularization, and resembles the well-known heuristic e-scaling.
Using a follow the regularized leader analysis and novel bounds for this iterative and proximal regularized setting, we bound the complexity of our algorithm as O(n2/e3). However, this result depends on two still unproved assumptions, that we leave for future work.
We have not analyzed the experimental performance of the algorithm, but we believe it should be equivalent to e-scaling. This strategy has been shown to outperform the standard approach of solving only one scaling problem with very small regularization.
Moreover, one of the advantages of our algorithm and e-scaling is that they are any-time algorithms, producing better solutions as more computational time is available.
@inproceedings{ortiz2019proximal,
title = {A proximal formulation of regularized optimal transport},
booktitle = {Research Report. Supervisor: Gergely Neu. University Pompeu Fabra (UPF), Barcelona},
author = {Ortiz-Haro, Joaquim},
pdf = {/assets/ortiz2019proximal.pdf},
institution = {University Pompeu Fabra, Barcelona},
year = {2019}
}
Joaquim Ortiz-Haro.
Master Thesis. Supervisors: Nicolas Mansard and Justin Carpentier. LAAS-CNRS, Toulouse, 2019.
|
|
Legged robots do not require a flat and smooth surface to advance and have, therefore, thepotential to work in unstructured and complex environments. These robots, like humansand animals, can climb stairs and ladders and cross challenging terrain, which is essentialfor exploration and rescue applications.However, walking is a complex task. The robot movement must be a consequence of creat-ing contacts with the environment. Every time a new contact is made the dynamics becomediscontinuous. Also, legged robots are unstable and subject to many constraints, which re-stricts the potential movements. Finally, contact planning has a combinatorial structure: therobot chooses the optimal subset of contacts out of an infinite number of possible contactpoints.Trajectory optimization in robotics is often decoupled into two independent modules. Aproblem formulation, that models the robot motion and transforms it into a mathematicaloptimization problem, and the resolution of this optimization problem with a suitable op-timization solver. We rather envision this question as a two-way interaction between thetwo modules, that can not be considered separately. Indeed, the robotics problem formu-lation must also be a consequence of the understanding of the capabilities of the numericaloptimization algorithm.We propose a simultaneous trajectory and contact optimization with an augmented La-grangian algorithm. The contacts with the environment are modeled in a continuous waywith complementarity constraints. For example, the normal contact model is defined bytwo inequalities: positive force and distance to the surface, and a complementarity relation:either the force or the distance to the surface is zero.These constraints are added as path constraints in a continuous optimal control formula-tion. Using a direct collocation, the formulation is converted into a non-linear constrainedoptimization problem. Optimization problems with complementarity constraints have a de-generated and challenging structure and do not fulfil basic constraints qualifications usedby standard numerical solvers. Therefore, we propose to solve it using an augmented La-grangian algorithm, which offers a good behaviour under this type of constraints and canbe efficiently warmstarted.In this thesis we present our preliminary results with a robot manipulator interacting withthe environment. Optimizing simultaneously trajectory and contacts, we compute interest-ing motions such as multiple surface touching and pick and place tasks. We also outline ourfuture plans to generate walking motions with legged robots.
@inproceedings{ortizdeharo:hal-02180282,
title = {Simultaneous trajectory and contact optimization with an augmented Lagrangian algorithm},
booktitle = {Master Thesis. Supervisors: Nicolas Mansard and Justin Carpentier. LAAS-CNRS, Toulouse},
author = {Ortiz-Haro, Joaquim},
hal_local_reference = {Rapport LAAS n{\textdegree} 19208},
year = {2019},
pdf = {https://hal.laas.fr/hal-02180282/file/2019_msc_quim_ortiz.pdf},
hal_id = {hal-02180282},
hal_version = {v1}
}
Joaquim Ortiz-Haro.
Bachelor Thesis. Supervisor: Joan Sola. IRI (Robotics Institute) and UPC, Barcelona, 2017.
|
|
An event camera has independent pixels that sends information, called “events”
when they perceive a local change of brightness. The information is transmitted
asynchronously exactly when the change occurs, with a microsecond resolution,
making this sensor suitable for fast robotics applications.
We present two new tracking and mapping algorithms, designed to work in
parallel to estimate the 6 DOF (Degrees Of Freedom) trajectory and the structure
of the scene in line based environments.
The tracking thread is based on a Landmark Based map and an asynchronous EKF
(Extended Kalman Filter) filter to estimate event per event the state of the camera
unlocking the true potential of the camera.
Inside the mapping thread, a line extraction algorithm has been designed to find
3D segments in the Point cloud, computed using event – ray tracing into a
discretized world.
Both algorithms have been built from scratch, and at this moment, only tested
independently in simulation.
We have obtained very good results on three synthetic self-made datasets.
Some pieces of the complete Parallel Tracking and Mapping system are still
missing. The current good work and results encourages to improve and finish the
algorithm to achieve the implementation on the real event based camera.
@inproceedings{ortiz2017parallel,
title = {Parallel tracking and mapping algorithms for an event based camera},
booktitle = {Bachelor Thesis. Supervisor: Joan Sola. IRI (Robotics Institute) and UPC, Barcelona},
institution = {Robotics Institute IRI and UPC},
author = {Ortiz-Haro, Joaquim},
pdf = {https://upcommons.upc.edu/handle/2117/111945},
year = {2017}
}
Wolfgang Hönig, Joaquim Ortiz-Haro, and Toussaint Marc.
Motion planning workshop, in IROS, 2022.
|
|
We consider time-optimal motion planning for dynamical systems that are translation-invariant, a property that holds for many mobile robots, such as differential-drives, cars, airplanes, and multirotors. Previous benchmarks have typically focused on comparing approaches within the same algorithmic class, eg, sampling-based approaches may be benchmarked using the open motion planning library (OMPL). We provide the first benchmark that compares search-, sampling-, and optimization-based time-optimal motion planning on multiple dynamical systems in different settings.
@inproceedings{honig2022benchmarking,
title = {Benchmarking Sampling-, Search-, and Optimization-based Approaches for Time-Optimal Kinodynamic Mobile Robot Motion Planning},
author = {Hönig, Wolfgang and Ortiz-Haro, Joaquim and Marc, Toussaint},
year = {2022},
booktitle = {Motion planning workshop, in IROS},
pdf = {https://motion-planning-workshop.kavrakilab.org/papers/kinodynamic-motion-planning-benchmark.pdf}
}
Welf Rehberg, Joaquim Ortiz-Haro, Marc Toussaint, and Wolfgang Hönig.
2023.
|
@misc{rehberg2023comparison,
title = {Comparison of Optimization-Based Methods for Energy-Optimal Quadrotor Motion Planning},
author = {Rehberg, Welf and Ortiz-Haro, Joaquim and Toussaint, Marc and Hönig, Wolfgang},
year = {2023},
eprint = {2304.14062},
archiveprefix = {arXiv},
primaryclass = {cs.RO},
pdf = {https://arxiv.org/abs/2304.14062}
}
Student Supervision
Efficient Kinodynamic Motion Planning with Reinforcement Learning Policies
Alexander Weingart. Co-supervision: Joaquim Ortiz-Haro and Wolfgang Hoenig. Master Thesis Computer Science (TU-Berlin).
2023.
Neural Scene Representations for Sequential Reasoning