Qi An

EAAI Journal 2026 Journal Article

An Adaptive Image Dehazing Network with Multi-Color Feature for Complex Real-World Hazy Scenes

• Zhiyu Lyu
• Qi An
• Yan Chen

Real-world hazy scenes can be broadly categorized into four types based on haze distribution and concentration: light homogeneous haze, dense homogeneous haze, light non-homogeneous haze, and dense non-homogeneous haze. However, many existing dehazing models are tailored for specific haze types, struggling to generalize effectively across these diverse conditions. Additionally, these models commonly extract feature information in the Red, Green, and Blue (RGB) color space, which makes it challenging to extract sufficient feature information in various hazy scenes. To address this issue, we propose an Adaptive Network (AdaNet) for multiple hazy scenes. The network includes two sub-networks: a color-guided feature extraction network and a scene reconstruction network. The color-guided feature extraction network is used to capture sufficient color, detail, and other feature information in both RGB and Luminance, Chroma Red, Chroma Blue (YCrCb) color spaces. For light and dense non-homogeneous hazy scenes, we enhance the scene reconstruction network with the Feature Selection Units (FSU) to filter out less relevant information, ensuring precise recovery of critical local details. Additionally, to tackle dehazing in light and dense homogeneous hazy scenes, we integrate the Feature Fusion Units (FFU) that combine multi-level features to improve overall feature utilization. Extensive experiments on multiple datasets with diverse hazy scenes demonstrate that our AdaNet outperforms state-of-the-art dehazing models, producing high-quality dehazed images in quadruple haze scenarios and ensuring reliability for high-level visual tasks in real-world hazy scenes.

ICRA Conference 2025 Conference Paper

Integrated Motion State Prediction for Sit-to-Stand and Stand-to-Sit Motions Toward Effective Power Assist Control

• Kai Ren
• Yuichi Nakamura 0001
• Kazuaki Kondo
• Kei Shimonishi
• Takahide Ito
• Jun-ichiro Furukawa
• Qi An

Sit-to-stand and stand-to-sit motions are important in daily activities. However, elderly individuals often find these motions difficult to perform with declining lower limb strength, which causes a considerable reduction to their quality of life. In this study, a sensing method for controlling robotic assistive devices was proposed. This method utilizes electromyographic measurements and a deep neural network to predict motion initiation, and it estimates the timing of triggering assistive devices. Experimental results indicate that four muscle synergy patterns are required to represent the sit-to-stand and stand-to-sit motions together, with two of them being shared between both movements. Subsequently, a long short-term memory network was designed to forecast these two motions, and the result indicates that the prediction accuracy reached 92. 95% ± 0. 83% with forecasting time of 300 ms.

IROS Conference 2025 Conference Paper

Mapping in Indoor Environments Including Transparent Objects Using Stereo Polarization Camera and Projector

• Yusuke Ogihara
• Hiroshi Higuchi
• Takuya Igaue
• Qi An
• Atsushi Yamashita

This paper proposes a method for generating maps in indoor environments that include transparent objects by using a stereo polarization camera and projector. Conventional sensors like LiDAR and stereo cameras struggle with glass, as they rely on diffuse reflection, while glass allows light to pass through. In contrast, polarization cameras can measure light polarization and estimate surface normals, enabling depth estimation by combining polarization and RGB information. However, when measuring transparent objects, reflected and transmitted light cancel each other out, reducing polarization contrast, and the RGB information causes the depth estimation to output the depth of objects behind the glass. To address this issue, this paper proposes a novel method that (1) improves the S/N ration in polarization measument via diffuse reflection on non-glass regions and (2) masks out the RGB color from polarimetric depth estimation to not compute depth map of objects behind the glass to obtain depth images that include glass surfaces. Additionally, (3) in the mapping part, depth estimation is repeated at multiple locations, and the results are integrated using self-localization to generate a complete environmental map. Experiments in an indoor environment confirmed the effectiveness of the proposed method, enabling glass-inclusive depth estimation and successful map generation on a mobile robot.

ICML Conference 2025 Conference Paper

Quantifying Memory Utilization with Effective State-Size

• Rom N. Parnichkun
• Neehal Tumma
• Armin W. Thomas
• Alessandro Moro
• Qi An
• Taiji Suzuki
• Atsushi Yamashita
• Michael Poli

As the space of causal sequence modeling architectures continues to grow, the need to develop a general framework for their analysis becomes increasingly important. With this aim, we draw insights from classical signal processing and control theory, to develop a quantitative measure of memory utilization: the internal mechanisms through which a model stores past information to produce future outputs. This metric, which we call effective state-size (ESS), is tailored to the fundamental class of systems with input-invariant and input-varying linear operators, encompassing a variety of computational units such as variants of attention, convolutions, and recurrences. Unlike prior work on memory utilization, which either relies on raw operator visualizations (e. g. attention maps), or simply the total memory capacity (i. e. cache size) of a model, our metrics provide highly interpretable and actionable measurements. In particular, we show how ESS can be leveraged to improve initialization strategies, inform novel regularizers and advance the performance-efficiency frontier through model distillation. Furthermore, we demonstrate that the effect of context delimiters (such as end-of-speech tokens) on ESS highlights cross-architectural differences in how large language models utilize their available memory to recall information. Overall, we find that ESS provides valuable insights into the dynamics that dictate memory utilization, enabling the design of more efficient and effective sequence models.

ICML Conference 2024 Conference Paper

State-Free Inference of State-Space Models: The *Transfer Function* Approach

• Rom N. Parnichkun
• Stefano Massaroli
• Alessandro Moro
• Jimmy T. H. Smith
• Ramin M. Hasani
• Mathias Lechner
• Qi An
• Christopher Ré

We approach designing a state-space model for deep learning applications through its dual representation, the transfer function, and uncover a highly efficient sequence parallel inference algorithm that is state-free: unlike other proposed algorithms, state-free inference does not incur any significant memory or computational cost with an increase in state size. We achieve this using properties of the proposed frequency domain transfer function parametrization, which enables direct computation of its corresponding convolutional kernel’s spectrum via a single Fast Fourier Transform. Our experimental results across multiple sequence lengths and state sizes illustrates, on average, a 35% training speed improvement over S4 layers – parametrized in time-domain – on the Long Range Arena benchmark, while delivering state-of-the-art downstream performances over other attention-free approaches. Moreover, we report improved perplexity in language modeling over a long convolutional Hyena baseline, by simply introducing our transfer function parametrization. Our code is available at https: //github. com/ruke1ire/RTF.

IROS Conference 2023 Conference Paper

Risk-Sensitive Mobile Robot Navigation in Crowded Environment via Offline Reinforcement Learning

• Jiaxu Wu
• Yusheng Wang 0001
• Hajime Asama
• Qi An
• Atsushi Yamashita

Mobile robot navigation in a human-populated environment has been of great interest to the research community in recent years, referred to as crowd navigation. Currently, offline reinforcement learning (RL)-based method has been introduced to this domain, for its ability to alleviate the sim2real gap brought by online RL which relies on simulators to execute training, and its scalability to use the same dataset to train for differently customized rewards. However, the performance of the navigation policy suffered from the distributional shift between the training data and the input during deployment, since when it gets an input out of the training data distribution, the learned policy has the risk of choosing an erroneous action that leads to catastrophic failure such as colliding with a human. To realize risk sensitivity and improve the safety of the offline RL agent during deployment, this work proposes a multipolicy control framework that combines offline RL navigation policy with a risk detector and a force-based risk-avoiding policy. In particular, a Lyapunov density model is learned using the latent feature of the offline RL policy and works as a risk detector to switch the control to the risk-avoiding policy when the robot has a tendency to go out of the area supported by the training data. Experimental results showed that the proposed method was able to learn navigation in a crowded scene from the offline trajectory dataset and the risk detector substantially reduces the collision rate of the vanilla offline RL agent while maintaining the navigation efficiency outperforming the state-of-the-art methods.

IROS Conference 2020 Conference Paper

Development of dementia care training system based on augmented reality and whole body wearable tactile sensor

• Tomoki Hiramatsu
• Masaya Kamei
• Daiji Inoue
• Akihiro Kawamura
• Qi An
• Ryo Kurazume

This study develops a training system for a multimodal comprehensive care methodology for dementia patients called Humanitude. Humanitude has attracted much attention as a gentle and effective care technique. It consists of four main techniques, namely, eye contact, verbal communication, touch, and standing up, and more than 150 care elements. Learning Humanitude thus requires much time. To provide an effective training system for Humanitude, we develop a training system that realizes sensing and interaction simultaneously by combining a real entity and augmented reality technology. To imitate the interaction between a patient and a caregiver, we superimpose a three-dimensional CG model of a patient’s face onto the head of a soft doll using augmented reality technology. Touch information such as position and force is sensed using the whole body wearable tactile sensor developed to quantify touch skills. This training system enables the evaluation of eye contact and touch skills simultaneously. We build a prototype of the proposed training system and evaluate the usefulness of the system in public lectures.

ICRA Conference 2015 Conference Paper

Analysis of muscle synergy contribution on human standing-up motion using a neuro-musculoskeletal model

• Qi An
• Yuki Ishikawa
• Shinya Aoi
• Tetsuro Funato
• Hiroyuki Oka
• Hiroshi Yamakawa
• Atsushi Yamashita
• Hajime Asama

It is important to understand the mechanism of human standing-up motion to improve the declined physical ability of the elderly people. This study employs the concept of muscle synergies (modular structure of coordinative muscle activation) to understand how humans coordinate their muscles to achieve the standing-up motion. Neuro-musculoskeletal model was developed to represent human body to generate standing-up motion. Using the developed model, forward dynamic simulation was used to analyze how humans utilized the muscle synergies to realize the motion. Results showed that the developed model could generate the standing-up motion with four muscle synergies rather than controlling individual muscles. Moreover, further analysis showed that three different strategies of the standing-up motion could be generated only by changing the start time of the particular muscle synergy.

ICRA Conference 2014 Conference Paper

Generation of human standing-up motion with muscle synergies using forward dynamic simulation

• Qi An
• Yuki Ishikawa
• Tetsuro Funato
• Shinya Aoi
• Hiroyuki Oka
• Hiroshi Yamakawa
• Atsushi Yamashita
• Hajime Asama

The standing-up motion is one of the most important activities of daily livings. In order to understand the strategy to achieve the standing-up motion, muscle synergy analysis is applied to the measured data during human standing-up motion. In addition, musculoskeletal model which consists of three body segments and nine muscles in lower limb is developed to ensure that the standing-up motion can be generated by muscle synergies. As a result, three muscle synergies have been extracted from the human standing-up motion, and each synergy strongly corresponded to characteristic kinematic events: momentum flexion, momentum transfer, and posture stabilization. Results of forward dynamic simulation show that the standing-up motion can be achieved by controlling time-varying weighting coefficient of three muscle synergies instead of controlling individual nine muscles.

ICML Conference 2008 Conference Paper

Hierarchical kernel stick-breaking process for multi-task image analysis

• Qi An
• Chunping Wang 0001
• Ivo Shterev
• Eric Wang
• Lawrence Carin
• David B. Dunson