arXiv subjects

Stelian Coros

Publications and source records attributed to Stelian Coros.

At least 55 records · Page 3Linked to original sources

arXiv2025

Spatio-Temporal Motion Retargeting for Quadruped Robots

This work presents a motion retargeting approach for legged robots, aimed at transferring the dynamic and agile movements to robots from source motions. In particular, we guide the imitation learning procedures by transferring motions from source to target, effectively bridging the morphological disparities while ensuring the physical feasibility of the target system. In the first stage, we focus on motion retargeting at the kinematic level by generating kinematically feasible whole-body motions from keypoint trajectories. Following this, we refine the motion at the dynamic level by adjusting it in the temporal domain while adhering to physical constraints. This process facilitates policy training via reinforcement learning, enabling precise and robust motion tracking. We demonstrate that our approach successfully transforms noisy motion sources, such as hand-held camera videos, into robot-specific motions that align with the morphology and physical properties of the target robots. Moreover, we demonstrate terrain-aware motion retargeting to perform BackFlip on top of a box. We successfully deployed these skills to four robots with different dimensions and physical properties in the real world through hardware experiments.

cs.RO↗

arXiv2025

Self-supervised Learning of Latent Space Dynamics

Modeling the dynamic behavior of deformable objects is crucial for creating realistic digital worlds. While conventional simulations produce high-quality motions, their computational costs are often prohibitive. Subspace simulation techniques address this challenge by restricting deformations to a lower-dimensional space, improving performance while maintaining visually compelling results. However, even subspace methods struggle to meet the stringent performance demands of portable devices such as virtual reality headsets and mobile platforms. To overcome this limitation, we introduce a novel subspace simulation framework powered by a neural latent-space integrator. Our approach leverages self-supervised learning to enhance inference stability and generalization. By operating entirely within latent space, our method eliminates the need for full-space computations, resulting in a highly efficient method well-suited for deployment on portable devices. We demonstrate the effectiveness of our approach on challenging examples involving rods, shells, and solids, showcasing its versatility and potential for widespread adoption.

cs.GR↗

arXiv2025

Epistemically-guided forward-backward exploration

Zero-shot reinforcement learning is necessary for extracting optimal policies in absence of concrete rewards for fast adaptation to future problem settings. Forward-backward representations (FB) have emerged as a promising method for learning optimal policies in absence of rewards via a factorization of the policy occupancy measure. However, up until now, FB and many similar zero-shot reinforcement learning algorithms have been decoupled from the exploration problem, generally relying on other exploration algorithms for data collection. We argue that FB representations should fundamentally be used for exploration in order to learn more efficiently. With this goal in mind, we design exploration policies that arise naturally from the FB representation that minimize the posterior variance of the FB representation, hence minimizing its epistemic uncertainty. We empirically demonstrate that such principled exploration strategies improve sample complexity of the FB algorithm considerably in comparison to other exploration methods. Code is publicly available at https://sites.google.com/view/fbee-url.

cs.LG↗

arXiv2025

Budget-optimal multi-robot layout design for box sorting

Robotic systems are routinely used in the logistics industry to enhance operational efficiency, but the design of robot workspaces remains a complex and manual task, which limits the system's flexibility to changing demands. This paper aims to automate robot workspace design by proposing a computational framework to generate a budget-minimizing layout by selectively placing stationary robots on a floor grid to sort packages from given input and output locations. Finding a good layout that minimizes the hardware budget while ensuring motion feasibility is a challenging combinatorial problem with nonconvex motion constraints. We propose a new optimization-based approach that models layout planning as a subgraph optimization problem subject to network flow constraints. Our core insight is to abstract away motion constraints from the layout optimization by precomputing a kinematic reachability graph and then extract the optimal layout on this ground graph. We validate the motion feasibility of our approach by proposing a simple task assignment and motion planning technique. We benchmark our algorithm on problems with various grid resolutions and number of outputs and show improvements in memory efficiency over a heuristic search algorithm.

cs.RO↗

arXiv2025

A comparison of visual representations for real-world reinforcement learning in the context of vacuum gripping

When manipulating objects in the real world, we need reactive feedback policies that take into account sensor information to inform decisions. This study aims to determine how different encoders can be used in a reinforcement learning (RL) framework to interpret the spatial environment in the local surroundings of a robot arm. Our investigation focuses on comparing real-world vision with 3D scene inputs, exploring new architectures in the process. We built on the SERL framework, providing us with a sample efficient and stable RL foundation we could build upon, while keeping training times minimal. The results of this study indicate that spatial information helps to significantly outperform the visual counterpart, tested on a box picking task with a vacuum gripper. The code and videos of the evaluations are available at https://github.com/nisutte/voxel-serl.

cs.RO↗

arXiv2025

NeoRL: Efficient Exploration for Nonepisodic RL

We study the problem of nonepisodic reinforcement learning (RL) for nonlinear dynamical systems, where the system dynamics are unknown and the RL agent has to learn from a single trajectory, i.e., without resets. We propose Nonepisodic Optimistic RL (NeoRL), an approach based on the principle of optimism in the face of uncertainty. NeoRL uses well-calibrated probabilistic models and plans optimistically w.r.t. the epistemic uncertainty about the unknown dynamics. Under continuity and bounded energy assumptions on the system, we provide a first-of-its-kind regret bound of $O(Γ_T \sqrt{T})$ for general nonlinear systems with Gaussian process dynamics. We compare NeoRL to other baselines on several deep RL environments and empirically demonstrate that NeoRL achieves the optimal average cost while incurring the least regret.

cs.LG↗

arXiv2025

PokeFlex: A Real-World Dataset of Volumetric Deformable Objects for Robotics

Data-driven methods have shown great potential in solving challenging manipulation tasks; however, their application in the domain of deformable objects has been constrained, in part, by the lack of data. To address this lack, we propose PokeFlex, a dataset featuring real-world multimodal data that is paired and annotated. The modalities include 3D textured meshes, point clouds, RGB images, and depth maps. Such data can be leveraged for several downstream tasks, such as online 3D mesh reconstruction, and it can potentially enable underexplored applications such as the real-world deployment of traditional control methods based on mesh simulations. To deal with the challenges posed by real-world 3D mesh reconstruction, we leverage a professional volumetric capture system that allows complete 360° reconstruction. PokeFlex consists of 18 deformable objects with varying stiffness and shapes. Deformations are generated by dropping objects onto a flat surface or by poking the objects with a robot arm. Interaction wrenches and contact locations are also reported for the latter case. Using different data modalities, we demonstrated a use case for our dataset training models that, given the novelty of the multimodal nature of Pokeflex, constitute the state-of-the-art in multi-object online template-based mesh reconstruction from multimodal data, to the best of our knowledge. We refer the reader to our website ( https://pokeflex-dataset.github.io/ ) for further demos and examples.

cs.RO↗

arXiv2024

RobotKeyframing: Learning Locomotion with High-Level Objectives via Mixture of Dense and Sparse Rewards

This paper presents a novel learning-based control framework that uses keyframing to incorporate high-level objectives in natural locomotion for legged robots. These high-level objectives are specified as a variable number of partial or complete pose targets that are spaced arbitrarily in time. Our proposed framework utilizes a multi-critic reinforcement learning algorithm to effectively handle the mixture of dense and sparse rewards. Additionally, it employs a transformer-based encoder to accommodate a variable number of input targets, each associated with specific time-to-arrivals. Throughout simulation and hardware experiments, we demonstrate that our framework can effectively satisfy the target keyframe sequence at the required times. In the experiments, the multi-critic method significantly reduces the effort of hyperparameter tuning compared to the standard single-critic alternative. Moreover, the proposed transformer-based architecture enables robots to anticipate future goals, which results in quantitative improvements in their ability to reach their targets.

cs.RO↗

arXiv2024

MetaLoco: Universal Quadrupedal Locomotion with Meta-Reinforcement Learning and Motion Imitation

This work presents a meta-reinforcement learning approach to develop a universal locomotion control policy capable of zero-shot generalization across diverse quadrupedal platforms. The proposed method trains an RL agent equipped with a memory unit to imitate reference motions using a small set of procedurally generated quadruped robots. Through comprehensive simulation and real-world hardware experiments, we demonstrate the efficacy of our approach in achieving locomotion across various robots without requiring robot-specific fine-tuning. Furthermore, we highlight the critical role of the memory unit in enabling generalization, facilitating rapid adaptation to changes in the robot properties, and improving sample efficiency.

cs.RO↗

arXiv2024

RL + Model-based Control: Using On-demand Optimal Control to Learn Versatile Legged Locomotion

This paper presents a control framework that combines model-based optimal control and reinforcement learning (RL) to achieve versatile and robust legged locomotion. Our approach enhances the RL training process by incorporating on-demand reference motions generated through finite-horizon optimal control, covering a broad range of velocities and gaits. These reference motions serve as targets for the RL policy to imitate, leading to the development of robust control policies that can be learned with reliability. Furthermore, by utilizing realistic simulation data that captures whole-body dynamics, RL effectively overcomes the inherent limitations in reference motions imposed by modeling simplifications. We validate the robustness and controllability of the RL training process within our framework through a series of experiments. In these experiments, our method showcases its capability to generalize reference motions and effectively handle more complex locomotion tasks that may pose challenges for the simplified model, thanks to RL's flexibility. Additionally, our framework effortlessly supports the training of control policies for robots with diverse dimensions, eliminating the necessity for robot-specific adjustments in the reward function and hyperparameters.

cs.RO↗

arXiv2024

PokeFlex: Towards a Real-World Dataset of Deformable Objects for Robotic Manipulation

Advancing robotic manipulation of deformable objects can enable automation of repetitive tasks across multiple industries, from food processing to textiles and healthcare. Yet robots struggle with the high dimensionality of deformable objects and their complex dynamics. While data-driven methods have shown potential for solving manipulation tasks, their application in the domain of deformable objects has been constrained by the lack of data. To address this, we propose PokeFlex, a pilot dataset featuring real-world 3D mesh data of actively deformed objects, together with the corresponding forces and torques applied by a robotic arm, using a simple poking strategy. Deformations are captured with a professional volumetric capture system that allows for complete 360-degree reconstruction. The PokeFlex dataset consists of five deformable objects with varying stiffness and shapes. Additionally, we leverage the PokeFlex dataset to train a vision model for online 3D mesh reconstruction from a single image and a template mesh. We refer readers to the supplementary material and to our website ( https://pokeflex-dataset.github.io/ ) for demos and examples of our dataset.

cs.RO↗

arXiv2024

Bridging the Sim-to-Real Gap with Bayesian Inference

We present SIM-FSVGD for learning robot dynamics from data. As opposed to traditional methods, SIM-FSVGD leverages low-fidelity physical priors, e.g., in the form of simulators, to regularize the training of neural network models. While learning accurate dynamics already in the low data regime, SIM-FSVGD scales and excels also when more data is available. We empirically show that learning with implicit physical priors results in accurate mean model estimation as well as precise uncertainty quantification. We demonstrate the effectiveness of SIM-FSVGD in bridging the sim-to-real gap on a high-performance RC racecar system. Using model-based RL, we demonstrate a highly dynamic parking maneuver with drifting, using less than half the data compared to the state of the art.

cs.RO↗

arXiv2024

Rethinking Robustness Assessment: Adversarial Attacks on Learning-based Quadrupedal Locomotion Controllers

Legged locomotion has recently achieved remarkable success with the progress of machine learning techniques, especially deep reinforcement learning (RL). Controllers employing neural networks have demonstrated empirical and qualitative robustness against real-world uncertainties, including sensor noise and external perturbations. However, formally investigating the vulnerabilities of these locomotion controllers remains a challenge. This difficulty arises from the requirement to pinpoint vulnerabilities across a long-tailed distribution within a high-dimensional, temporally sequential space. As a first step towards quantitative verification, we propose a computational method that leverages sequential adversarial attacks to identify weaknesses in learned locomotion controllers. Our research demonstrates that, even state-of-the-art robust controllers can fail significantly under well-designed, low-magnitude adversarial sequence. Through experiments in simulation and on the real robot, we validate our approach's effectiveness, and we illustrate how the results it generates can be used to robustify the original policy and offer valuable insights into the safety of these black-box policies. Project page: https://fanshi14.github.io/me/rss24.html

cs.RO↗

arXiv2024

TRTM: Template-based Reconstruction and Target-oriented Manipulation of Crumpled Cloths

Precise reconstruction and manipulation of the crumpled cloths is challenging due to the high dimensionality of cloth models, as well as the limited observation at self-occluded regions. We leverage the recent progress in the field of single-view human reconstruction to template-based reconstruct crumpled cloths from their top-view depth observations only, with our proposed sim-real registration protocols. In contrast to previous implicit cloth representations, our reconstruction mesh explicitly describes the positions and visibilities of the entire cloth mesh vertices, enabling more efficient dual-arm and single-arm target-oriented manipulations. Experiments demonstrate that our TRTM system can be applied to daily cloths that have similar topologies as our template mesh, but with different shapes, sizes, patterns, and physical properties. Videos, datasets, pre-trained models, and code can be downloaded from our project website: https://wenbwa.github.io/TRTM/ .

cs.RO↗

arXiv2024

Modal Folding: Discovering Smooth Folding Patterns for Sheet Materials using Strain-Space Modes

Folding can transform mundane objects such as napkins into stunning works of art. However, finding new folding transformations for sheet materials is a challenging problem that requires expertise and real-world experimentation. In this paper, we present Modal Folding -- an automated approach for discovering energetically optimal folding transformations, i.e., large deformations that require little mechanical work. For small deformations, minimizing internal energy for fixed displacement magnitudes leads to the well-known elastic eigenmodes. While linear modes provide promising directions for bending, they cannot capture the rotational motion required for folding. To overcome this limitation, we introduce strain-space modes -- nonlinear analogues of elastic eigenmodes that operate on per-element curvatures instead of vertices. Using strain-space modes to determine target curvatures for bending elements, we can generate complex nonlinear folding motions by simply minimizing the sheet's internal energy. Our modal folding approach offers a systematic and automated way to create complex designs. We demonstrate the effectiveness of our method with simulation results for a range of shapes and materials, and validate our designs with physical prototypes.

cs.GR↗

arXiv2024

Differentiable Geodesic Distance for Intrinsic Minimization on Triangle Meshes

Computing intrinsic distances on discrete surfaces is at the heart of many minimization problems in geometry processing and beyond. Solving these problems is extremely challenging as it demands the computation of on-surface distances along with their derivatives. We present a novel approach for intrinsic minimization of distance-based objectives defined on triangle meshes. Using a variational formulation of shortest-path geodesics, we compute first and second-order distance derivatives based on the implicit function theorem, thus opening the door to efficient Newton-type minimization solvers. We demonstrate our differentiable geodesic distance framework on a wide range of examples, including geodesic networks and membranes on surfaces of arbitrary genus, two-way coupling between hosting surface and embedded system, differentiable geodesic Voronoi diagrams, and efficient computation of Karcher means on complex shapes. Our analysis shows that second-order descent methods based on our differentiable geodesics outperform existing first-order and quasi-Newton methods by large margins.

cs.GR↗

arXiv2024

Differentiable Voronoi Diagrams for Simulation of Cell-Based Mechanical Systems

Navigating topological transitions in cellular mechanical systems is a significant challenge for existing simulation methods. While abstract models lack predictive capabilities at the cellular level, explicit network representations struggle with topology changes, and per-cell representations are computationally too demanding for large-scale simulations. To address these challenges, we propose a novel cell-centered approach based on differentiable Voronoi diagrams. Representing each cell with a Voronoi site, our method defines shape and topology of the interface network implicitly. In this way, we substantially reduce the number of problem variables, eliminate the need for explicit contact handling, and ensure continuous geometry changes during topological transitions. Closed-form derivatives of network positions facilitate simulation with Newton-type methods for a wide range of per-cell energies. Finally, we extend our differentiable Voronoi diagrams to enable coupling with arbitrary rigid and deformable boundaries. We apply our approach to a diverse set of examples, highlighting splitting and merging of cells as well as neighborhood changes. We illustrate applications to inverse problems by matching soap foam simulations to real-world images. Comparative analysis with explicit cell models reveals that our method achieves qualitatively comparable results at significantly faster computation times.

cs.GR↗

arXiv2024

Neural Modes: Self-supervised Learning of Nonlinear Modal Subspaces

We propose a self-supervised approach for learning physics-based subspaces for real-time simulation. Existing learning-based methods construct subspaces by approximating pre-defined simulation data in a purely geometric way. However, this approach tends to produce high-energy configurations, leads to entangled latent space dimensions, and generalizes poorly beyond the training set. To overcome these limitations, we propose a self-supervised approach that directly minimizes the system's mechanical energy during training. We show that our method leads to learned subspaces that reflect physical equilibrium constraints, resolve overfitting issues of previous methods, and offer interpretable latent space parameters.

cs.LG↗