Cristina Mata

Yoo Sung Jang, Kanchana Ranasinghe, Cristina Mata et al.

Cross-embodiment learning from human demonstrations is hindered by the visual gap between human and robot embodiments. While self-supervised learning (SSL) backbones encode rich inter-class semantics of general objects, we show they fail to establish correspondence between human and robot hands. We propose LACE, a framework that aligns human and robot visual representations in the latent space of these backbones by leveraging correspondences between shared body parts across embodiments as sparse supervision. These annotations can be automatically obtained via forward kinematics, and single robot demonstration is sufficient to train the model. Our semantic alignment loss matches distributions incurred by corresponding features, lifting patch-level supervision to semantic-level alignment, while a Gram loss preserves pretrained feature quality. This alignment enables robot policies to leverage abundant human data when robot demonstrations are scarce: in zero-shot transfer, policies using LACE-DINO outperform those using DINO by a large margin (65\%), with consistent gains in low-data regimes and out-of-distribution environments.

Cristina Mata, Kanchana Ranasinghe, Michael S. Ryoo

Unsupervised domain adaptation (UDA) involves learning class semantics from labeled data within a source domain that generalize to an unseen target domain. UDA methods are particularly impactful for semantic segmentation, where annotations are more difficult to collect than in image classification. Despite recent advances in large-scale vision-language representation learning, UDA methods for segmentation have not taken advantage of the domain-agnostic properties of text. To address this, we present a novel Covariance-based Pixel-Text loss, CoPT, that uses domain-agnostic text embeddings to learn domain-invariant features in an image segmentation encoder. The text embeddings are generated through our LLM Domain Template process, where an LLM is used to generate source and target domain descriptions that are fed to a frozen CLIP model and combined. In experiments on four benchmarks we show that a model trained using CoPT achieves the new state of the art performance on UDA for segmentation. The code can be found at https://github.com/cfmata/CoPT.

Xiang Li, Cristina Mata, Jongwoo Park et al.

Vision Language Models (VLMs) have recently been leveraged to generate robotic actions, forming Vision-Language-Action (VLA) models. However, directly adapting a pretrained VLM for robotic control remains challenging, particularly when constrained by a limited number of robot demonstrations. In this work, we introduce LLaRA: Large Language and Robotics Assistant, a framework that formulates robot action policy as visuo-textual conversations and enables an efficient transfer of a pretrained VLM into a powerful VLA, motivated by the success of visual instruction tuning in Computer Vision. First, we present an automated pipeline to generate conversation-style instruction tuning data for robots from existing behavior cloning datasets, aligning robotic actions with image pixel coordinates. Further, we enhance this dataset in a self-supervised manner by defining six auxiliary tasks, without requiring any additional action annotations. We show that a VLM finetuned with a limited amount of such datasets can produce meaningful action decisions for robotic control. Through experiments across multiple simulated and real-world tasks, we demonstrate that LLaRA achieves state-of-the-art performance while preserving the generalization capabilities of large language models. The code, datasets, and pretrained models are available at https://github.com/LostXine/LLaRA.

Kanchana Ranasinghe, Xiang Li, E-Ro Nguyen et al.

We present LangToMo, a vision-language-action framework structured as a dual-system architecture that uses pixel motion forecasts as intermediate representations. Our high-level System 2, an image diffusion model, generates text-conditioned pixel motion sequences from a single frame to guide robot control. Pixel motion-a universal, interpretable, and motion-centric representation-can be extracted from videos in a weakly-supervised manner, enabling diffusion model training on any video-caption data. Treating generated pixel motion as learned universal representations, our low level System 1 module translates these into robot actions via motion-to-action mapping functions, which can be either hand-crafted or learned with minimal supervision. System 2 operates as a high-level policy applied at sparse temporal intervals, while System 1 acts as a low-level policy at dense temporal intervals. This hierarchical decoupling enables flexible, scalable, and generalizable robot control under both unsupervised and supervised settings, bridging the gap between language, motion, and action. Checkout https://kahnchana.github.io/LangToMo

Cristina Mata, Michael S. Ryoo, Henrik Turbell

Semantic segmentation relies on many dense pixel-wise annotations to achieve the best performance, but owing to the difficulty of obtaining accurate annotations for real world data, practitioners train on large-scale synthetic datasets. Unpaired image translation is one method used to address the ensuing domain gap by generating more realistic training data in low-data regimes. Current methods for unpaired image translation train generative adversarial networks (GANs) to perform the translation and enforce pixel-level semantic matching through cycle consistency. These methods do not guarantee that the semantic matching holds, posing a problem for semantic segmentation where performance is sensitive to noisy pixel labels. We propose a novel image translation method, Domain Adversarial Kernel Prediction Network (DA-KPN), that guarantees semantic matching between the synthetic label and translation. DA-KPN estimates pixel-wise input transformation parameters of a lightweight and simple translation function. To ensure the pixel-wise transformation is realistic, DA-KPN uses multi-scale discriminators to distinguish between translated and target samples. We show DA-KPN outperforms previous GAN-based methods on syn2real benchmarks for semantic segmentation with limited access to real image labels and achieves comparable performance on face parsing.

Cristina Mata, Nick Locascio, Mohammed Azeem Sheikh et al.

Autonomous checkout systems rely on visual and sensory inputs to carry out fine-grained scene understanding in retail environments. Retail environments present unique challenges compared to typical indoor scenes owing to the vast number of densely packed, unique yet similar objects. The problem becomes even more difficult when only RGB input is available, especially for data-hungry tasks such as instance segmentation. To address the lack of datasets for retail, we present StandardSim, a large-scale photorealistic synthetic dataset featuring annotations for semantic segmentation, instance segmentation, depth estimation, and object detection. Our dataset provides multiple views per scene, enabling multi-view representation learning. Further, we introduce a novel task central to autonomous checkout called change detection, requiring pixel-level classification of takes, puts and shifts in objects over time. We benchmark widely-used models for segmentation and depth estimation on our dataset, show that our test set constitutes a difficult benchmark compared to current smaller-scale datasets and that our training set provides models with crucial information for autonomous checkout tasks.

Cristina Mata, Guy Ben-Yosef, Boris Katz

Many deep learning architectures for semantic segmentation involve a Fully Convolutional Neural Network (FCN) followed by a Conditional Random Field (CRF) to carry out inference over an image. These models typically involve unary potentials based on local appearance features computed by FCNs, and binary potentials based on the displacement between pixels. We show that while current methods succeed in segmenting whole objects, they perform poorly in situations involving a large number of object parts. We therefore suggest incorporating into the inference algorithm additional higher-order potentials inspired by the way humans identify and localize parts. We incorporate two relations that were shown to be useful to human object identification - containment and attachment - into the energy term of the CRF and evaluate their performance on the Pascal VOC Parts dataset. Our experimental results show that the segmentation of fine parts is positively affected by the addition of these two relations, and that the segmentation of fine parts can be further influenced by complex structural features.