arXiv subjects

Cristian Romero

Publications and source records attributed to Cristian Romero.

4 recordsLinked to original sources

arXiv2026

DCGrasp: Distance-aware Controllable Grasp Generation

Generating 3D hand-object interactions is essential for applications in robotics, XR, and synthetic data generation, where flexible controllability and strong generalization to diverse object geometries are required. However, existing methods rarely satisfy these requirements, limiting their practical applicability. We present DCGrasp, a distance-aware controllable grasp generation system built on a novel grasp energy term. This term computes Distance Profile, a signed distance from each hand vertex to the nearest object point, coupled with distance-aware weighting, effectively capturing the semantically similar hand-object interaction in near-contact regions while remaining invariant to object and hand identity. Given various controllable signals, DCGrasp first generates a Distance Profile based on a Diffusion Transformer, together with a corresponding candidate hand pose. We then refine the candidate pose through optimization, enforcing consistency between the optimized hand pose and the generated Distance Profile in near-contact regions. Our experiments show that DCGrasp produces high-quality, physically plausible grasps with flexible user control, generalizing to diverse object and hand shapes and scales. Our work establishes a robust and versatile pipeline for the synthesis of controllable 3D hand-object interactions.

cs.CV↗

arXiv2023

FLSH -- Friendly Library for the Simulation of Humans

Computer models of humans are ubiquitous throughout computer animation and computer vision. However, these models rarely represent the dynamics of human motion, as this requires adding a complex layer that solves body motion in response to external interactions and according to the laws of physics. FLSH is a library that facilitates this task for researchers and developers who are not interested in the nuisances of physics simulation, but want to easily integrate dynamic humans in their applications. FLSH provides easy access to three flavors of body physics, with different features and computational complexity: skeletal dynamics, full soft-tissue dynamics, and reduced-order modeling of soft-tissue dynamics. In all three cases, the simulation models are built on top of the pseudo-standard SMPL parametric body model.

cs.GR↗

arXiv2023

Empirical Contrast Model for High-Contrast Imaging -- A VLT/SPHERE Case Study

The ability to accurately predict the contrast achieved from high contrast imagers is important for efficient scheduling and quality control measures in modern observatories. We aim to consistently predict and measure the raw contrast achieved by SPHERE/IRDIS on a frame by frame basis to improve the efficiency and scientific yield with SPHERE at the Very Large Telescope (VLT).Contrast curves were calculated for over 5 years of archival data using the most common SPHERE/IRDIS coronagraphic mode in the H2/H3 dual band filter, consisting of approximately 80,000 individual frames. These were merged and interpolated with atmospheric data to create a large data-base of contrast curves with associated features. An empirical power law model for contrast, motivated by physical considerations, was then trained and finally tested on an out-of-sample test data set. At an angular separation of 300 mas, the contrast model achieved a mean (out-of-sample) test error of 0.13 magnitudes with the residual 5-95% percentiles between -0.23 and 0.64 magnitude respectively. The models test set root mean square error (RMSE) between 250-600 mas was between 0.31 - 0.40 magnitudes which is equivalent with other state-of-the-art contrast models presented in the literature. In general, the model performed best for targets between 5-9 G-band magnitude, with degraded performance for targets outside this range. This model is currently being incorporated into the Paranal SCUBA software for first level quality control and real time scheduling support.

astro-ph.IM↗

arXiv2023

Data-Free Learning of Reduced-Order Kinematics

Physical systems ranging from elastic bodies to kinematic linkages are defined on high-dimensional configuration spaces, yet their typical low-energy configurations are concentrated on much lower-dimensional subspaces. This work addresses the challenge of identifying such subspaces automatically: given as input an energy function for a high-dimensional system, we produce a low-dimensional map whose image parameterizes a diverse yet low-energy submanifold of configurations. The only additional input needed is a single seed configuration for the system to initialize our procedure; no dataset of trajectories is required. We represent subspaces as neural networks that map a low-dimensional latent vector to the full configuration space, and propose a training scheme to fit network parameters to any system of interest. This formulation is effective across a very general range of physical systems; our experiments demonstrate not only nonlinear and very low-dimensional elastic body and cloth subspaces, but also more general systems like colliding rigid bodies and linkages. We briefly explore applications built on this formulation, including manipulation, latent interpolation, and sampling.

cs.GR↗