CRAFT: Adapting VLA Models to Contact-rich Manipulation via Force-aware Curriculum Fine-tuning

Vision-Language-Action (VLA) models have shown a strong capability in enabling robots to execute general instructions, yet they struggle with contact-rich manipulation tasks, where success requires precise alignment, stable contact maintenance, and effective handling of deformable objects. A fundamental challenge arises from the imbalance between high-entropy vision and language inputs and low-entropy but critical force signals, which often leads to over-reliance on perception and unstable control. To address this, we introduce CRAFT, a force-aware curriculum fine-tuning framework that integrates a variational information bottleneck module to regulate vision and language embeddings during early training. This curriculum strategy encourages the model to prioritize force signals initially, before progressively restoring access to the full multimodal information. To enable force-aware learning, we further design a homologous leader-follower teleoperation system that collects synchronized vision, language, and force data across diverse contact-rich tasks. Real-world experiments demonstrate that CRAFT consistently improves task success, generalizes to unseen objects and novel task variations, and adapts effectively across diverse VLA architectures, enabling robust and generalizable contact-rich manipulation.

A comprehensive framework for occluded human pose estimation

Occlusion presents a significant challenge in human pose estimation. The challenges posed by occlusion can be attributed to the following factors: 1) Data: The collection and annotation of occluded human pose samples are relatively challenging. 2) Feature: Occlusion can cause feature confusion due to the high similarity between the target person and interfering individuals. 3) Inference: Robust inference becomes challenging due to the loss of complete body structural information. The existing methods designed for occluded human pose estimation usually focus on addressing only one of these factors. In this paper, we propose a comprehensive framework DAG (Data, Attention, Graph) to address the performance degradation caused by occlusion. Specifically, we introduce the mask joints with instance paste data augmentation technique to simulate occlusion scenarios. Additionally, an Adaptive Discriminative Attention Module (ADAM) is proposed to effectively enhance the features of target individuals. Furthermore, we present the Feature-Guided Multi-Hop GCN (FGMP-GCN) to fully explore the prior knowledge of body structure and improve pose estimation results. Through extensive experiments conducted on three benchmark datasets for occluded human pose estimation, we demonstrate that the proposed method outperforms existing methods. Code and data will be publicly available.

Validating CircaCP: a Generic Sleep-Wake Cycle Detection Algorithm

Sleep-wake cycle detection is a key step when extrapolating sleep patterns from actigraphy data. Numerous supervised detection algorithms have been developed with parameters estimated from and optimized for a particular dataset, yet their generalizability from sensor to sensor or study to study is unknown. In this paper, we propose and validate an unsupervised algorithm -- CircaCP -- to detect sleep-wake cycles from minute-by-minute actigraphy data. It first uses a robust cosinor model to estimate circadian rhythm, then searches for a single change point (CP) within each cycle. We used CircaCP to estimate sleep/wake onset times (S/WOTs) from 2125 indviduals' data in the MESA Sleep study and compared the estimated S/WOTs against self-reported S/WOT event markers. Lastly, we quantified the biases between estimated and self-reported S/WOTs, as well as variation in S/WOTs contributed by the two methods, using linear mixed-effects models and variance component analysis. On average, SOTs estimated by CircaCP were five minutes behind those reported by event markers, and WOTs estimated by CircaCP were less than one minute behind those reported by markers. These differences accounted for less than 0.2% variability in SOTs and in WOTs, taking into account other sources of between-subject variations. By focusing on the commonality in human circadian rhythms captured by actigraphy, our algorithm transferred seamlessly from hip-worn ActiGraph data collected from children in our previous study to wrist-worn Actiwatch data collected from adults. The large between- and within-subject variability highlights the need for estimating individual-level S/WOTs when conducting actigraphy research. The generalizability of our algorithm also suggests that it could be widely applied to actigraphy data collected by other wearable sensors.