H. Jin Kim

Sungha Kim, Gawon Lee, Jusuk Lee et al.

Maximum entropy reinforcement learning (MaxEnt-RL) enables robust exploration, yet practical implementations often restrict policies to simple Gaussians. While recent approaches incorporate expressive generative policies via importance-weighted supervised learning, they are prone to importance weight collapse, which limits their scalability in high-dimensional action spaces. Our key insight is to mitigate this limitation by localizing the sampling region, avoiding the weight degeneracy induced by importance sampling over the entire action space. To instantiate this insight, we introduce \textbf{FLAG} (\textbf{F}low policy with \textbf{L}atent-\textbf{A}ugmented \textbf{G}uidance). FLAG augments the state space with a flow latent variable and optimizes a provably consistent proxy MaxEnt-RL objective. We empirically demonstrate that FLAG enables expressive policy optimization with limited importance samples and scales to high-dimensional control tasks. Furthermore, FLAG achieves state-of-the-art performance across challenging benchmarks. Our project webpage: https://flag-rl.github.io/

Jusuk Lee, Seungjae Lee, Jonghun Shin et al.

Robot manipulation critically depends on perception that preserves the action-relevant aspects of a scene. Yet most robot learning pipelines are built upon visual encoders pre-trained for static recognition or vision-language alignment, leaving motion understanding to downstream policies. We introduce DynaFLIP, a dynamics-aware multimodal pre-training framework that pushes motion understanding upstream into perception. We construct image-language-3D flow triplets from heterogeneous human and robot videos, and use these triplets as training-time supervision to shape an image-only encoder. Our key idea is to encourage the three modalities to span a small simplex volume in the shared hyperspherical space -- a smaller simplex volume indicating stronger alignment. To avoid the geometric ambiguity and trivial collapse of naive volume minimization, we combine simplex-volume minimization with a cosine regularizer and a contrastive objective. Our analyses show that DynaFLIP focuses on control-relevant regions critical for manipulation. The resulting dynamics-aware representations serve as reusable visual backbones and consistently outperform baselines across diverse downstream policies, including VLAs. We validate this across diverse simulation and real-world setups, with gains reaching +22.5% under out-of-distribution scenarios. Our results suggest that robot generalization improves when visual representations are trained to encode not just what is present, but how the world changes under action.

Jonghae Park, Daesol Cho, Jusuk Lee et al.

Unsupervised skill discovery in reinforcement learning (RL) aims to learn diverse behaviors without relying on external rewards. However, current methods often overlook the periodic nature of learned skills, focusing instead on increasing the mutual dependence between states and skills or maximizing the distance traveled in latent space. Considering that many robotic tasks - particularly those involving locomotion - require periodic behaviors across varying timescales, the ability to discover diverse periodic skills is essential. Motivated by this, we propose Periodic Skill Discovery (PSD), a framework that discovers periodic behaviors in an unsupervised manner. The key idea of PSD is to train an encoder that maps states to a circular latent space, thereby naturally encoding periodicity in the latent representation. By capturing temporal distance, PSD can effectively learn skills with diverse periods in complex robotic tasks, even with pixel-based observations. We further show that these learned skills achieve high performance on downstream tasks such as hurdling. Moreover, integrating PSD with an existing skill discovery method offers more diverse behaviors, thus broadening the agent's repertoire. Our code and demos are available at https://jonghaepark.github.io/psd/

Taekbeom Lee, Youngseok Jang, H. Jin Kim

Existence of symmetric objects, whose observation at different viewpoints can be identical, can deteriorate the performance of simultaneous localization and mapping(SLAM). This work proposes a system for robustly optimizing the pose of cameras and objects even in the presence of symmetric objects. We classify objects into three categories depending on their symmetry characteristics, which is efficient and effective in that it allows to deal with general objects and the objects in the same category can be associated with the same type of ambiguity. Then we extract only the unambiguous parameters corresponding to each category and use them in data association and joint optimization of the camera and object pose. The proposed approach provides significant robustness to the SLAM performance by removing the ambiguous parameters and utilizing as much useful geometric information as possible. Comparison with baseline algorithms confirms the superior performance of the proposed system in terms of object tracking and pose estimation, even in challenging scenarios where the baseline fails.

Dongseok Shim, H. Jin Kim

Monocular depth estimation plays a critical role in various computer vision and robotics applications such as localization, mapping, and 3D object detection. Recently, learning-based algorithms achieve huge success in depth estimation by training models with a large amount of data in a supervised manner. However, it is challenging to acquire dense ground truth depth labels for supervised training, and the unsupervised depth estimation using monocular sequences emerges as a promising alternative. Unfortunately, most studies on unsupervised depth estimation explore loss functions or occlusion masks, and there is little change in model architecture in that ConvNet-based encoder-decoder structure becomes a de-facto standard for depth estimation. In this paper, we employ a convolution-free Swin Transformer as an image feature extractor so that the network can capture both local geometric features and global semantic features for depth estimation. Also, we propose a Densely Cascaded Multi-scale Network (DCMNet) that connects every feature map directly with another from different scales via a top-down cascade pathway. This densely cascaded connectivity reinforces the interconnection between decoding layers and produces high-quality multi-scale depth outputs. The experiments on two different datasets, KITTI and Make3D, demonstrate that our proposed method outperforms existing state-of-the-art unsupervised algorithms.

Dongseok Shim, Seungjae Lee, H. Jin Kim

As previous representations for reinforcement learning cannot effectively incorporate a human-intuitive understanding of the 3D environment, they usually suffer from sub-optimal performances. In this paper, we present Semantic-aware Neural Radiance Fields for Reinforcement Learning (SNeRL), which jointly optimizes semantic-aware neural radiance fields (NeRF) with a convolutional encoder to learn 3D-aware neural implicit representation from multi-view images. We introduce 3D semantic and distilled feature fields in parallel to the RGB radiance fields in NeRF to learn semantic and object-centric representation for reinforcement learning. SNeRL outperforms not only previous pixel-based representations but also recent 3D-aware representations both in model-free and model-based reinforcement learning.

Seungjae Lee, Jigang Kim, Inkyu Jang et al.

Hierarchical Reinforcement Learning (HRL) has made notable progress in complex control tasks by leveraging temporal abstraction. However, previous HRL algorithms often suffer from serious data inefficiency as environments get large. The extended components, $i.e.$, goal space and length of episodes, impose a burden on either one or both high-level and low-level policies since both levels share the total horizon of the episode. In this paper, we present a method of Decoupling Horizons Using a Graph in Hierarchical Reinforcement Learning (DHRL) which can alleviate this problem by decoupling the horizons of high-level and low-level policies and bridging the gap between the length of both horizons using a graph. DHRL provides a freely stretchable high-level action interval, which facilitates longer temporal abstraction and faster training in complex tasks. Our method outperforms state-of-the-art HRL algorithms in typical HRL environments. Moreover, DHRL achieves long and complex locomotion and manipulation tasks.

Daesol Cho, Jigang Kim, H. Jin Kim

Current reinforcement learning (RL) in robotics often experiences difficulty in generalizing to new downstream tasks due to the innate task-specific training paradigm. To alleviate it, unsupervised RL, a framework that pre-trains the agent in a task-agnostic manner without access to the task-specific reward, leverages active exploration for distilling diverse experience into essential skills or reusable knowledge. For exploiting such benefits also in robotic manipulation, we propose an unsupervised method for transferable manipulation skill discovery that ties structured exploration toward interacting behavior and transferable skill learning. It not only enables the agent to learn interaction behavior, the key aspect of the robotic manipulation learning, without access to the environment reward, but also to generalize to arbitrary downstream manipulation tasks with the learned task-agnostic skills. Through comparative experiments, we show that our approach achieves the most diverse interacting behavior and significantly improves sample efficiency in downstream tasks including the extension to multi-object, multitask problems.

Daesol Cho, Seungjae Lee, H. Jin Kim

Current reinforcement learning (RL) often suffers when solving a challenging exploration problem where the desired outcomes or high rewards are rarely observed. Even though curriculum RL, a framework that solves complex tasks by proposing a sequence of surrogate tasks, shows reasonable results, most of the previous works still have difficulty in proposing curriculum due to the absence of a mechanism for obtaining calibrated guidance to the desired outcome state without any prior domain knowledge. To alleviate it, we propose an uncertainty & temporal distance-aware curriculum goal generation method for the outcome-directed RL via solving a bipartite matching problem. It could not only provide precisely calibrated guidance of the curriculum to the desired outcome states but also bring much better sample efficiency and geometry-agnostic curriculum goal proposal capability compared to previous curriculum RL methods. We demonstrate that our algorithm significantly outperforms these prior methods in a variety of challenging navigation tasks and robotic manipulation tasks in a quantitative and qualitative way.

Jigang Kim, J. hyeon Park, Daesol Cho et al.

Deep reinforcement learning has enabled robots to learn motor skills from environmental interactions with minimal to no prior knowledge. However, existing reinforcement learning algorithms assume an episodic setting, in which the agent resets to a fixed initial state distribution at the end of each episode, to successfully train the agents from repeated trials. Such reset mechanism, while trivial for simulated tasks, can be challenging to provide for real-world robotics tasks. Resets in robotic systems often require extensive human supervision and task-specific workarounds, which contradicts the goal of autonomous robot learning. In this paper, we propose an extension to conventional reinforcement learning towards greater autonomy by introducing an additional agent that learns to reset in a self-supervised manner. The reset agent preemptively triggers a reset to prevent manual resets and implicitly imposes a curriculum for the forward agent. We apply our method to learn from scratch on a suite of simulated and real-world continuous control tasks and demonstrate that the reset agent successfully learns to reduce manual resets whilst also allowing the forward policy to improve gradually over time.

Jigang Kim, Dohyun Jang, H. Jin Kim

Deploying multiple robots for target search and tracking has many practical applications, yet the challenge of planning over unknown or partially known targets remains difficult to address. With recent advances in deep learning, intelligent control techniques such as reinforcement learning have enabled agents to learn autonomously from environment interactions with little to no prior knowledge. Such methods can address the exploration-exploitation tradeoff of planning over unknown targets in a data-driven manner, eliminating the reliance on heuristics typical of traditional approaches and streamlining the decision-making pipeline with end-to-end training. In this paper, we propose a multi-agent reinforcement learning technique with target map building based on distributed Gaussian process. We leverage the distributed Gaussian process to encode belief over the target locations and efficiently plan over unknown targets. We evaluate the performance and transferability of the trained policy in simulation and demonstrate the method on a swarm of micro unmanned aerial vehicles with hardware experiments.

Jeongjun Choi, Dongseok Shim, H. Jin Kim

Thanks to the development of 2D keypoint detectors, monocular 3D human pose estimation (HPE) via 2D-to-3D uplifting approaches have achieved remarkable improvements. Still, monocular 3D HPE is a challenging problem due to the inherent depth ambiguities and occlusions. To handle this problem, many previous works exploit temporal information to mitigate such difficulties. However, there are many real-world applications where frame sequences are not accessible. This paper focuses on reconstructing a 3D pose from a single 2D keypoint detection. Rather than exploiting temporal information, we alleviate the depth ambiguity by generating multiple 3D pose candidates which can be mapped to an identical 2D keypoint. We build a novel diffusion-based framework to effectively sample diverse 3D poses from an off-the-shelf 2D detector. By considering the correlation between human joints by replacing the conventional denoising U-Net with graph convolutional network, our approach accomplishes further performance improvements. We evaluate our method on the widely adopted Human3.6M and HumanEva-I datasets. Comprehensive experiments are conducted to prove the efficacy of the proposed method, and they confirm that our model outperforms state-of-the-art multi-hypothesis 3D HPE methods.

Daesol Cho, Dongseok Shim, H. Jin Kim

Offline reinforcement learning (Offline RL) suffers from the innate distributional shift as it cannot interact with the physical environment during training. To alleviate such limitation, state-based offline RL leverages a learned dynamics model from the logged experience and augments the predicted state transition to extend the data distribution. For exploiting such benefit also on the image-based RL, we firstly propose a generative model, S2P (State2Pixel), which synthesizes the raw pixel of the agent from its corresponding state. It enables bridging the gap between the state and the image domain in RL algorithms, and virtually exploring unseen image distribution via model-based transition in the state space. Through experiments, we confirm that our S2P-based image synthesis not only improves the image-based offline RL performance but also shows powerful generalization capability on unseen tasks.

Seungjae Lee, Daesol Cho, Jonghae Park et al.

Recent curriculum Reinforcement Learning (RL) has shown notable progress in solving complex tasks by proposing sequences of surrogate tasks. However, the previous approaches often face challenges when they generate curriculum goals in a high-dimensional space. Thus, they usually rely on manually specified goal spaces. To alleviate this limitation and improve the scalability of the curriculum, we propose a novel curriculum method that automatically defines the semantic goal space which contains vital information for the curriculum process, and suggests curriculum goals over it. To define the semantic goal space, our method discretizes continuous observations via vector quantized-variational autoencoders (VQ-VAE) and restores the temporal relations between the discretized observations by a graph. Concurrently, ours suggests uncertainty and temporal distance-aware curriculum goals that converges to the final goals over the automatically composed goal space. We demonstrate that the proposed method allows efficient explorations in an uninformed environment with raw goal examples only. Also, ours outperforms the state-of-the-art curriculum RL methods on data efficiency and performance, in various goal-reaching tasks even with ego-centric visual inputs.

Daesol Cho, Seungjae Lee, H. Jin Kim

Reinforcement learning (RL) often faces the challenges of uninformed search problems where the agent should explore without access to the domain knowledge such as characteristics of the environment or external rewards. To tackle these challenges, this work proposes a new approach for curriculum RL called Diversify for Disagreement & Conquer (D2C). Unlike previous curriculum learning methods, D2C requires only a few examples of desired outcomes and works in any environment, regardless of its geometry or the distribution of the desired outcome examples. The proposed method performs diversification of the goal-conditional classifiers to identify similarities between visited and desired outcome states and ensures that the classifiers disagree on states from out-of-distribution, which enables quantifying the unexplored region and designing an arbitrary goal-conditioned intrinsic reward signal in a simple and intuitive way. The proposed method then employs bipartite matching to define a curriculum learning objective that produces a sequence of well-adjusted intermediate goals, which enable the agent to automatically explore and conquer the unexplored region. We present experimental results demonstrating that D2C outperforms prior curriculum RL methods in both quantitative and qualitative aspects, even with the arbitrarily distributed desired outcome examples.

Lin Yang, Jinwoo Lee, Domenico Campolo et al.

Aerial manipulation combines the maneuverability of multirotors with the dexterity of robotic arms to perform complex tasks in cluttered spaces. Yet planning safe, dynamically feasible trajectories remains difficult due to whole-body collision avoidance and the conservativeness of common geometric abstractions such as bounding boxes or ellipsoids. We present a whole-body motion planning and safety-critical control framework for aerial manipulators built on superquadrics (SQs). Using an SQ-plus-proxy representation, we model both the vehicle and obstacles with differentiable, geometry-accurate surfaces. Leveraging this representation, we introduce a maximum-clearance planner that fuses Voronoi diagrams with an equilibrium-manifold formulation to generate smooth, collision-aware trajectories. We further design a safety-critical controller that jointly enforces thrust limits and collision avoidance via high-order control barrier functions. In simulation, our approach outperforms sampling-based planners in cluttered environments, producing faster, safer, and smoother trajectories and exceeding ellipsoid-based baselines in geometric fidelity. Actual experiments on a physical aerial-manipulation platform confirm feasibility and robustness, demonstrating consistent performance across simulation and hardware settings. The video can be found at https://youtu.be/hQYKwrWf1Ak.

Byeongjun Kim, H. Jin Kim

We introduce Deep QP Safety Filter, a fully data-driven safety layer for black-box dynamical systems. Our method learns a Quadratic-Program (QP) safety filter without model knowledge by combining Hamilton-Jacobi (HJ) reachability with model-free learning. We construct contraction-based losses for both the safety value and its derivatives, and train two neural networks accordingly. In the exact setting, the learned critic converges to the viscosity solution (and its derivative), even for non-smooth values. Across diverse dynamical systems -- even including a hybrid system -- and multiple RL tasks, Deep QP Safety Filter substantially reduces pre-convergence failures while accelerating learning toward higher returns than strong baselines, offering a principled and practical route to safe, model-free control.

Taekbeom Lee, Dabin Kim, Youngseok Jang et al.

We present HERE, an active 3D scene reconstruction framework based on neural radiance fields, enabling high-fidelity implicit mapping. Our approach centers around an active learning strategy for camera trajectory generation, driven by accurate identification of unseen regions, which supports efficient data acquisition and precise scene reconstruction. The key to our approach is epistemic uncertainty quantification based on evidential deep learning, which directly captures data insufficiency and exhibits a strong correlation with reconstruction errors. This allows our framework to more reliably identify unexplored or poorly reconstructed regions compared to existing methods, leading to more informed and targeted exploration. Additionally, we design a hierarchical exploration strategy that leverages learned epistemic uncertainty, where local planning extracts target viewpoints from high-uncertainty voxels based on visibility for trajectory generation, and global planning uses uncertainty to guide large-scale coverage for efficient and comprehensive reconstruction. The effectiveness of the proposed method in active 3D reconstruction is demonstrated by achieving higher reconstruction completeness compared to previous approaches on photorealistic simulated scenes across varying scales, while a hardware demonstration further validates its real-world applicability.

Seungjae Lee, Yibin Wang, Haritheja Etukuru et al.

Generative modeling of complex behaviors from labeled datasets has been a longstanding problem in decision making. Unlike language or image generation, decision making requires modeling actions - continuous-valued vectors that are multimodal in their distribution, potentially drawn from uncurated sources, where generation errors can compound in sequential prediction. A recent class of models called Behavior Transformers (BeT) addresses this by discretizing actions using k-means clustering to capture different modes. However, k-means struggles to scale for high-dimensional action spaces or long sequences, and lacks gradient information, and thus BeT suffers in modeling long-range actions. In this work, we present Vector-Quantized Behavior Transformer (VQ-BeT), a versatile model for behavior generation that handles multimodal action prediction, conditional generation, and partial observations. VQ-BeT augments BeT by tokenizing continuous actions with a hierarchical vector quantization module. Across seven environments including simulated manipulation, autonomous driving, and robotics, VQ-BeT improves on state-of-the-art models such as BeT and Diffusion Policies. Importantly, we demonstrate VQ-BeT's improved ability to capture behavior modes while accelerating inference speed 5x over Diffusion Policies. Videos and code can be found https://sjlee.cc/vq-bet

Seungyeon Yoo, Youngseok Jang, Dabin Kim et al.

Visual navigation models often struggle in real-world dynamic environments due to limited robustness to the sim-to-real gap and the difficulty of training policies tailored to target deployment environments (e.g., households, restaurants, and factories). Although real-to-sim navigation simulation using 3D Gaussian Splatting (GS) can mitigate these challenges, prior GS-based works have considered only static scenes or non-photorealistic human obstacles built from simulator assets, despite the importance of safe navigation in dynamic environments. To address these issues, we propose ReaDy-Go, a novel real-to-sim simulation pipeline that synthesizes photorealistic dynamic scenarios in target environments by augmenting a reconstructed static GS scene with dynamic human GS obstacles, and trains navigation policies using the generated datasets. The pipeline provides three key contributions: (1) a dynamic GS simulator that integrates static scene GS with a human animation module, enabling the insertion of animatable human GS avatars and the synthesis of plausible human motions from 2D trajectories, (2) a navigation dataset generation framework that leverages the simulator along with a robot expert planner designed for dynamic GS representations and a human planner, and (3) robust navigation policies to both the sim-to-real gap and moving obstacles. The proposed simulator generates thousands of photorealistic navigation scenarios with animatable human GS avatars from arbitrary viewpoints. ReaDy-Go outperforms baselines across target environments in both simulation and real-world experiments, demonstrating improved navigation performance even after sim-to-real transfer and in the presence of moving obstacles. Moreover, zero-shot sim-to-real deployment in an unseen environment indicates its generalization potential. Project page: https://syeon-yoo.github.io/ready-go-site/.

Boseong Jeon, H. Jin Kim

This letter suggests an integrated approach for a drone (or multirotor) to perform an autonomous videography task in a 3-D obstacle environment by following a moving object. The proposed system includes 1) a target motion prediction module which can be applied to dense environments and 2) a hierarchical chasing planner based on a proposed metric for visibility. In the prediction module, we minimize observation error given that the target object itself does not collide with obstacles. The estimated future trajectory of target is obtained by covariant optimization. The other module, chasing planner, is in a bi-level structure composed of preplanner and smooth planner. In the first phase, we leverage a graph-search method to preplan a chasing corridor which incorporates safety and visibility of target during a time window. In the subsequent phase, we generate a smooth and dynamically feasible path within the corridor using quadratic programming (QP). We validate our approach with multiple complex scenarios and actual experiments. The source code can be found in https://github.com/icsl-Jeon/traj_gen_vis

Boseong Felipe Jeon, H. Jin Kim

This work deals with a moving target chasing mission of an aerial vehicle equipped with a vision sensor in a cluttered environment. In contrast to obstacle-free or sparse environments, the chaser should be able to handle collision and occlusion simultaneously with flight efficiency. In order to tackle these challenges with real-time replanning, we introduce a metric for target visibility and propose a cascaded chasing planner. By means of the graph-search methods, we first generate a sequence of chasing corridors and waypoints which ensure safety and optimize visibility. In the following phase, the corridors and waypoints are utilized as constraints and objective in quadratic programming from which we complete a dynamically feasible trajectory for chasing. The proposed algorithm is tested in multiple dense environments. The simulator AutoChaser with full code implementation and GUI can be found in https://github.com/icsl-Jeon/traj_gen_vis

Jeongjun Choi, Yeonsoo Park, H. Jin Kim

We present SceneNAT, a single-stage masked non-autoregressive Transformer that synthesizes complete 3D indoor scenes from natural language instructions through only a few parallel decoding passes, offering improved performance and efficiency compared to prior state-of-the-art approaches. SceneNAT is trained via masked modeling over fully discretized representations of both semantic and spatial attributes. By applying a masking strategy at both the attribute level and the instance level, the model can better capture intra-object and inter-object structure. To boost relational reasoning, SceneNAT employs a dedicated triplet predictor for modeling the scene's layout and object relationships by mapping a set of learnable relation queries to a sparse set of symbolic triplets (subject, predicate, object). Extensive experiments on the 3D-FRONT dataset demonstrate that SceneNAT achieves superior performance compared to state-of-the-art autoregressive and diffusion baselines in both semantic compliance and spatial arrangement accuracy, while operating with substantially lower computational cost.

Dahyun Oh, Minhyuk Yoon, H. Jin Kim

Cooperative multi-agent reinforcement learning (MARL) requires agents to discover joint strategies in a combinatorially large state-action space, yet effective coordination configurations are exceedingly rare. Intrinsic motivation, which augments task rewards with novelty bonuses, is a popular approach for driving exploration, but its effectiveness hinges on the exploration intensity $β$, where too large a value overwhelms the task signal and causes coordination collapse, while too small a value prevents discovery of rare strategies. We address two complementary challenges: adapting $β$ globally over training, and allocating the exploration budget across agents whose intrinsic reward signals vary in reliability. Our framework combines a return-conditioned sigmoid schedule (RCB) for global intensity control with a per-agent Reward Signal Quality (RSQ) metric that concentrates the exploration budget on agents with reliable signals. The core insight is that agents receiving noisy intrinsic rewards should explore less aggressively, and this allocation can be determined automatically from signal-to-noise statistics. Successor Distance (SD), a quasimetric intrinsic reward, naturally produces distinguishable per-agent signal quality, completing the framework with convergence and ordering preservation guarantees. On seven cooperative benchmarks (MPE, SMAX, MABrax), our method achieves top-tier returns across all environments.

Inkyu Jang, Chams E. Mballo, Claire J. Tomlin et al.

Stochastic control barrier functions (SCBFs) provide a safety-critical control framework for systems subject to stochastic disturbances by bounding the probability of remaining within a safe set. However, synthesizing a valid SCBF that explicitly reflects the true safety probability of the system, which is the most natural measure of safety, remains a challenge. This paper addresses this issue by adopting a spectral perspective, utilizing the linear operator that governs the evolution of the closed-loop system's safety probability. We find that the dominant eigenpair of this Koopman-like operator encodes fundamental safety information of the stochastic system. The dominant eigenfunction is a natural and valid SCBF, with values that explicitly quantify the relative long-term safety of the state, while the dominant eigenvalue indicates the global rate at which the safety probability decays. A practical synthesis algorithm is proposed, termed power-policy iteration, which jointly computes the dominant eigenpair and an optimized backup policy. The method is validated using simulation experiments on safety-critical dynamics models.

Changsuk Oh, Dongseok Shim, Taekbeom Lee et al.

Object removal refers to the process of erasing designated objects from an image while preserving the overall appearance, and it is one area where image inpainting is widely used in real-world applications. The performance of an object remover is quantitatively evaluated by measuring the quality of object removal results, similar to how the performance of an image inpainter is gauged. Current works reporting quantitative performance evaluations utilize original images as references. In this letter, to validate the current evaluation methods cannot properly evaluate the performance of an object remover, we create a dataset with object removal ground truth and compare the evaluations made by the current methods using original images to those utilizing object removal ground truth images. The disparities between two evaluation sets validate that the current methods are not suitable for measuring the performance of an object remover. Additionally, we propose new evaluation methods tailored to gauge the performance of an object remover. The proposed methods evaluate the performance through class-wise object removal results and utilize images without the target class objects as a comparison set. We confirm that the proposed methods can make judgments consistent with human evaluators in the COCO dataset, and that they can produce measurements aligning with those using object removal ground truth in the self-acquired dataset.

Dabin Kim, Inkyu Jang, Youngsoo Han et al.

Vision sensors are extensively used for localizing a robot's pose, particularly in environments where global localization tools such as GPS or motion capture systems are unavailable. In many visual navigation systems, localization is achieved by detecting and tracking visual features or landmarks, which provide information about the sensor's relative pose. For reliable feature tracking and accurate pose estimation, it is crucial to maintain visibility of a sufficient number of features. This requirement can sometimes conflict with the robot's overall task objective. In this paper, we approach it as a constrained control problem. By leveraging the invariance properties of visibility constraints within the robot's kinematic model, we propose a real-time safety filter based on quadratic programming. This filter takes a reference velocity command as input and produces a modified velocity that minimally deviates from the reference while ensuring the information score from the currently visible features remains above a user-specified threshold. Numerical simulations demonstrate that the proposed safety filter preserves the invariance condition and ensures the visibility of more features than the required minimum. We also validated its real-world performance by integrating it into a visual simultaneous localization and mapping (SLAM) algorithm, where it maintained high estimation quality in challenging environments, outperforming a simple tracking controller.

Dongseok Shim, Yichun Shi, Kejie Li et al.

Recent advancements in text-to-3D generation, building on the success of high-performance text-to-image generative models, have made it possible to create imaginative and richly textured 3D objects from textual descriptions. However, a key challenge remains in effectively decoupling light-independent and lighting-dependent components to enhance the quality of generated 3D models and their relighting performance. In this paper, we present MVLight, a novel light-conditioned multi-view diffusion model that explicitly integrates lighting conditions directly into the generation process. This enables the model to synthesize high-quality images that faithfully reflect the specified lighting environment across multiple camera views. By leveraging this capability to Score Distillation Sampling (SDS), we can effectively synthesize 3D models with improved geometric precision and relighting capabilities. We validate the effectiveness of MVLight through extensive experiments and a user study.

Hoseong Jung, Sungil Son, Daesol Cho et al.

Autonomous robotic systems should reason about resource control and its impact on subsequent maneuvers, especially when operating with limited energy budgets or restricted sensing. Learning-based control is effective in handling complex dynamics and represents the problem as a hybrid action space unifying discrete resource usage and continuous maneuvers. However, prior works on hybrid action space have not sufficiently captured the causal dependencies between resource usage and maneuvers. They have also overlooked the multi-modal nature of tactical decisions, both of which are critical in fast-evolving scenarios. In this paper, we propose TART, a Temporal Action Representation learning framework for Tactical resource control and subsequent maneuver generation. TART leverages contrastive learning based on a mutual information objective, designed to capture inherent temporal dependencies in resource-maneuver interactions. These learned representations are quantized into discrete codebook entries that condition the policy, capturing recurring tactical patterns and enabling multi-modal and temporally coherent behaviors. We evaluate TART in two domains where resource deployment is critical: (i) a maze navigation task where a limited budget of discrete actions provides enhanced mobility, and (ii) a high-fidelity air combat simulator in which an F-16 agent operates weapons and defensive systems in coordination with flight maneuvers. Across both domains, TART consistently outperforms hybrid-action baselines, demonstrating its effectiveness in leveraging limited resources and producing context-aware subsequent maneuvers.

Gawon Lee, Daesol Cho, H. Jin Kim

Multi-task reinforcement learning (MTRL) offers a promising approach to improve sample efficiency and generalization by training agents across multiple tasks, enabling knowledge sharing between them. However, applying MTRL to robotics remains challenging due to the high cost of collecting diverse task data. To address this, we propose MT-Lévy, a novel exploration strategy that enhances sample efficiency in MTRL environments by combining behavior sharing across tasks with temporally extended exploration inspired by Lévy flight. MT-Lévy leverages policies trained on related tasks to guide exploration towards key states, while dynamically adjusting exploration levels based on task success ratios. This approach enables more efficient state-space coverage, even in complex robotics environments. Empirical results demonstrate that MT-Lévy significantly improves exploration and sample efficiency, supported by quantitative and qualitative analyses. Ablation studies further highlight the contribution of each component, showing that combining behavior sharing with adaptive exploration strategies can significantly improve the practicality of MTRL in robotics applications.

Changhyun Choi, Sungha Kim, H. Jin Kim

Recently, it has been shown that investing computing resources in searching for good initial noise for a text-to-image diffusion model helps improve performance. However, previous studies required external models to evaluate the resulting images, which is impossible on GPUs with small VRAM. For these reasons, we apply Best-of-N inference-time scaling to algorithms that optimize the initial noise of a diffusion model without external models across multiple datasets and backbones. We demonstrate that inference-time scaling for text-to-image diffusion models in this setting quickly reaches a performance plateau, and a relatively small number of optimization steps suffices to achieve the maximum achievable performance with each algorithm.

Taekbeom Lee, Youngseok Jang, H. Jin Kim

Neural implicit representation has attracted attention in 3D reconstruction through various success cases. For further applications such as scene understanding or editing, several works have shown progress towards object compositional reconstruction. Despite their superior performance in observed regions, their performance is still limited in reconstructing objects that are partially observed. To better treat this problem, we introduce category-level neural fields that learn meaningful common 3D information among objects belonging to the same category present in the scene. Our key idea is to subcategorize objects based on their observed shape for better training of the category-level model. Then we take advantage of the neural field to conduct the challenging task of registering partially observed objects by selecting and aligning against representative objects selected by ray-based uncertainty. Experiments on both simulation and real-world datasets demonstrate that our method improves the reconstruction of unobserved parts for several categories.

Jigang Kim, Daesol Cho, H. Jin Kim

While reinforcement learning (RL) has achieved great success in acquiring complex skills solely from environmental interactions, it assumes that resets to the initial state are readily available at the end of each episode. Such an assumption hinders the autonomous learning of embodied agents due to the time-consuming and cumbersome workarounds for resetting in the physical world. Hence, there has been a growing interest in autonomous RL (ARL) methods that are capable of learning from non-episodic interactions. However, existing works on ARL are limited by their reliance on prior data and are unable to learn in environments where task-relevant interactions are sparse. In contrast, we propose a demonstration-free ARL algorithm via Implicit and Bi-directional Curriculum (IBC). With an auxiliary agent that is conditionally activated upon learning progress and a bidirectional goal curriculum based on optimal transport, our method outperforms previous methods, even the ones that leverage demonstrations.

Changsuk Oh, H. Jin Kim

The objective of the image inpainting task is to fill missing regions of an image in a visually plausible way. Recently, deep-learning-based image inpainting networks have generated outstanding results, and some utilize their models as object removers by masking unwanted objects in an image. However, while trying to better remove objects using their networks, the previous works pay less attention to the importance of the input mask. In this paper, we focus on generating the input mask to better remove objects using the off-the-shelf image inpainting network. We propose an automatic mask generator inspired by the explainable AI (XAI) method, whose output can better remove objects than a semantic segmentation mask. The proposed method generates an importance map using randomly sampled input masks and quantitatively estimated scores of the completed images obtained from the random masks. The output mask is selected by a judge module among the candidate masks which are generated from the importance map. We design the judge module to quantitatively estimate the quality of the object removal results. In addition, we empirically find that the evaluation methods used in the previous works reporting object removal results are not appropriate for estimating the performance of an object remover. Therefore, we propose new evaluation metrics (FID$^*$ and U-IDS$^*$) to properly evaluate the quality of object removers. Experiments confirm that our method shows better performance in removing target class objects than the masks generated from the semantic segmentation maps, and the two proposed metrics make judgments consistent with humans.

Dohyun Jang, Jaehyun Yoo, Clark Youngdong Son et al.

This paper proposes a cooperative environmental learning algorithm working in a fully distributed manner. A multi-robot system is more effective for exploration tasks than a single robot, but it involves the following challenges: 1) online distributed learning of environmental map using multiple robots; 2) generation of safe and efficient exploration path based on the learned map; and 3) maintenance of the scalability with respect to the number of robots. To this end, we divide the entire process into two stages of environmental learning and path planning. Distributed algorithms are applied in each stage and combined through communication between adjacent robots. The environmental learning algorithm uses a distributed Gaussian process, and the path planning algorithm uses a distributed Monte Carlo tree search. As a result, we build a scalable system without the constraint on the number of robots. Simulation results demonstrate the performance and scalability of the proposed system. Moreover, a real-world-dataset-based simulation validates the utility of our algorithm in a more realistic scenario.

Boseong Felipe Jeon, Changhyeon Kim, Hojoon Shin et al.

Rapidly generating an optimal chasing motion of a drone to follow a dynamic target among obstacles is challenging due to numerical issues rising from multiple conflicting objectives and non-convex constraints. This study proposes to resolve the difficulties with a fast and reliable pipeline that incorporates 1) a target movement forecaster and 2) a chasing planner. They are based on a sample-and-check approach that consists of the generation of high-quality candidate primitives and the feasibility tests with a light computation load. We forecast the movement of the target by selecting an optimal prediction among a set of candidates built from past observations. Based on the prediction, we construct a set of prospective chasing trajectories which reduce the high-order derivatives, while maintaining the desired relative distance from the predicted target movement. Then, the candidate trajectories are tested on safety of the chaser and visibility toward the target without loose approximation of the constraints. The proposed algorithm is thoroughly evaluated in challenging scenarios involving dynamic obstacles. Also, the overall process from the target recognition to the chasing motion planning is implemented fully onboard on a drone, demonstrating real-world applicability.

Jungwon Park, Dabin Kim, Gyeong Chan Kim et al.

This paper presents a new online multi-agent trajectory planning algorithm that guarantees to generate safe, dynamically feasible trajectories in a cluttered environment. The proposed algorithm utilizes a linear safe corridor (LSC) to formulate the distributed trajectory optimization problem with only feasible constraints, so it does not resort to slack variables or soft constraints to avoid optimization failure. We adopt a priority-based goal planning method to prevent the deadlock without an additional procedure to decide which robot to yield. The proposed algorithm can compute the trajectories for 60 agents on average 15.5 ms per agent with an Intel i7 laptop and shows a similar flight distance and distance compared to the baselines based on soft constraints. We verified that the proposed method can reach the goal without deadlock in both the random forest and the indoor space, and we validated the safety and operability of the proposed algorithm through a real flight test with ten quadrotors in a maze-like environment.

Dabin Kim, Gyeong Chan Kim, Youngseok Jang et al.

Visual navigation has been widely used for state estimation of micro aerial vehicles (MAVs). For stable visual navigation, MAVs should generate perception-aware paths which guarantee enough visible landmarks. Many previous works on perception-aware path planning focused on sampling-based planners. However, they may suffer from sample inefficiency, which leads to computational burden for finding a global optimal path. To address this issue, we suggest a perception-aware path planner which utilizes topological information of environments. Since the topological class of a path and visible landmarks during traveling the path are closely related, the proposed algorithm checks distinctive topological classes to choose the class with abundant visual information. Topological graph is extracted from the generalized Voronoi diagram of the environment and initial paths with different topological classes are found. To evaluate the perception quality of the classes, we divide the initial path into discrete segments where the points in each segment share similar visual information. The optimal class with high perception quality is selected, and a graph-based planner is utilized to generate path within the class. With simulations and real-world experiments, we confirmed that the proposed method could guarantee accurate visual navigation compared with the perception-agnostic method while showing improved computational efficiency than the sampling-based perception-aware planner.

Inkyu Jang, Dongjae Lee, Seungjae Lee et al.

Motion planners for mobile robots in unknown environments face the challenge of simultaneously maintaining both robustness against unmodeled uncertainties and persistent feasibility of the trajectory-finding problem. That is, while dealing with uncertainties, a motion planner must update its trajectory, adapting to the newly revealed environment in real-time; failing to do so may involve unsafe circumstances. Many existing planning algorithms guarantee these by maintaining the clearance needed to perform an emergency brake, which is itself a robust and persistently feasible maneuver. However, such maneuvers are not applicable for systems in which braking is impossible or risky, such as fixed-wing aircraft. To that end, we propose a real-time robust planner that recursively guarantees persistent feasibility without any need of braking. The planner ensures robustness against bounded uncertainties and persistent feasibility by constructing a loop of sequentially composed funnels, starting from the receding horizon local trajectory's forward reachable set. We implement the proposed algorithm for a robotic car tracking a speed-fixed reference trajectory. The experiment results show that the proposed algorithm can be run at faster than 16 Hz, while successfully keeping the system away from entering any dead-end, to maintain safety and feasibility.

Dongjae Lee, Inkyu Jang, Jeonghyun Byun et al.

Automation of excavation tasks requires real-time trajectory planning satisfying various constraints. To guarantee both constraint feasibility and real-time trajectory re-plannability, we present an integrated framework for real-time optimization-based trajectory planning of a hydraulic excavator. The proposed framework is composed of two main modules: a global planner and a real-time local planner. The global planner computes the entire global trajectory considering excavation volume and energy minimization while the local counterpart tracks the global trajectory in a receding horizon manner, satisfying dynamic feasibility, physical constraints, and disturbance-awareness. We validate the proposed planning algorithm in a simulation environment where two types of operations are conducted in the presence of emulated disturbance from hydraulic friction and soil-bucket interaction: shallow and deep excavation. The optimized global trajectories are obtained in an order of a second, which is tracked by the local planner at faster than 30 Hz. To the best of our knowledge, this work presents the first real-time motion planning framework that satisfies constraints of a hydraulic excavator, such as force/torque, power, cylinder displacement, and flow rate limits.

Jeonghyun Byun, Dongjae Lee, Hoseong Seo et al.

In this paper, an autonomous aerial manipulation task of pulling a plug out of an electric socket is conducted, where maintaining the stability and robustness is challenging due to sudden disappearance of a large interaction force. The abrupt change in the dynamical model before and after the separation of the plug can cause destabilization or mission failure. To accomplish aerial plug-pulling, we employ the concept of hybrid automata to divide the task into three operative modes, i.e, wire-pulling, stabilizing, and free-flight. Also, a strategy for trajectory generation and a design of disturbance-observer-based controllers for each operative mode are presented. Furthermore, the theory of hybrid automata is used to prove the stability and robustness during the mode transition. We validate the proposed trajectory generation and control method by an actual wire-pulling experiment with a multirotor-based aerial manipulator.

Dongseok Shim, H. Jin Kim

Deep neural networks have been widely studied in autonomous driving applications such as semantic segmentation or depth estimation. However, training a neural network in a supervised manner requires a large amount of annotated labels which are expensive and time-consuming to collect. Recent studies leverage synthetic data collected from a virtual environment which are much easier to acquire and more accurate compared to data from the real world, but they usually suffer from poor generalization due to the inherent domain shift problem. In this paper, we propose a Domain-Agnostic Contrastive Learning (DACL) which is a two-stage unsupervised domain adaptation framework with cyclic adversarial training and contrastive loss. DACL leads the neural network to learn domain-agnostic representation to overcome performance degradation when there exists a difference between training and test data distribution. Our proposed approach achieves better performance in the monocular depth estimation task compared to previous state-of-the-art methods and also shows effectiveness in the semantic segmentation task.

Dongseok Shim, H. Jin Kim

Previous studies on image classification have mainly focused on the performance of the networks, not on real-time operation or model compression. We propose a Gaussian Deep Recurrent visual Attention Model (GDRAM)- a reinforcement learning based lightweight deep neural network for large scale image classification that outperforms the conventional CNN (Convolutional Neural Network) which uses the entire image as input. Highly inspired by the biological visual recognition process, our model mimics the stochastic location of the retina with Gaussian distribution. We evaluate the model on Large cluttered MNIST, Large CIFAR-10 and Large CIFAR-100 datasets which are resized to 128 in both width and height.

Dongseok Shim, H. Jin Kim

Estimating a depth map from a single RGB image has been investigated widely for localization, mapping, and 3-dimensional object detection. Recent studies on a single-view depth estimation are mostly based on deep Convolutional neural Networks (ConvNets) which require a large amount of training data paired with densely annotated labels. Depth annotation tasks are both expensive and inefficient, so it is inevitable to leverage RGB images which can be collected very easily to boost the performance of ConvNets without depth labels. However, most self-supervised learning algorithms are focused on capturing the semantic information of images to improve the performance in classification or object detection, not in depth estimation. In this paper, we show that existing self-supervised methods do not perform well on depth estimation and propose a gradient-based self-supervised learning algorithm with momentum contrastive loss to help ConvNets extract the geometric information with unlabeled images. As a result, the network can estimate the depth map accurately with a relatively small amount of annotated data. To show that our method is independent of the model structure, we evaluate our method with two different monocular depth estimation algorithms. Our method outperforms the previous state-of-the-art self-supervised learning algorithms and shows the efficiency of labeled data in triple compared to random initialization on the NYU Depth v2 dataset.

Haram Kim, Pyojin Kim, H. Jin Kim

Detection of moving objects is an essential capability in dealing with dynamic environments. Most moving object detection algorithms have been designed for color images without depth. For robotic navigation where real-time RGB-D data is often readily available, utilization of the depth information would be beneficial for obstacle recognition. Here, we propose a simple moving object detection algorithm that uses RGB-D images. The proposed algorithm does not require estimating a background model. Instead, it uses an occlusion model which enables us to estimate the camera pose on a background confused with moving objects that dominate the scene. The proposed algorithm allows to separate the moving object detection and visual odometry (VO) so that an arbitrary robust VO method can be employed in a dynamic situation with a combination of moving object detection, whereas other VO algorithms for a dynamic environment are inseparable. In this paper, we use dense visual odometry (DVO) as a VO method with a bi-square regression weight. Experimental results show the segmentation accuracy and the performance improvement of DVO in the situations. We validate our algorithm in public datasets and our dataset which also publicly accessible.

Youngseok Jang, Hojoon Shin, H. Jin Kim

With the dominance of keyframe-based SLAM in the field of robotics, the relative frame poses between keyframes have typically been sacrificed for a faster algorithm to achieve online applications. However, those approaches can become insufficient for applications that may require refined poses of all frames, not just keyframes which are relatively sparse compared to all input frames. This paper proposes a novel algorithm to correct the relative frames between keyframes after the keyframes have been updated by a back-end optimization process. The correction model is derived using conservation of the measurement constraint between landmarks and the robot pose. The proposed algorithm is designed to be easily integrable to existing keyframe-based SLAM systems while exhibiting robust and accurate performance superior to existing interpolation methods. The algorithm also requires low computational resources and hence has a minimal burden on the whole SLAM pipeline. We provide the evaluation of the proposed pose correction algorithm in comparison to existing interpolation methods in various vector spaces, and our method has demonstrated excellent accuracy in both KITTI and EuRoC datasets.

Boseong Felipe Jeon, Dongseok Shim, H. Jin Kim

This work investigates an efficient trajectory generation for chasing a dynamic target, which incorporates the detectability objective. The proposed method actively guides the motion of a cinematographer drone so that the color of a target is well-distinguished against the colors of the background in the view of the drone. For the objective, we define a measure of color detectability given a chasing path. After computing a discrete path optimized for the metric, we generate a dynamically feasible trajectory. The whole pipeline can be updated on-the-fly to respond to the motion of the target. For the efficient discrete path generation, we construct a directed acyclic graph (DAG) for which a topological sorting can be determined analytically without the depth-first search. The smooth path is obtained in quadratic programming (QP) framework. We validate the enhanced performance of state-of-the-art object detection and tracking algorithms when the camera drone executes the trajectory obtained from the proposed method.

Dongjae Lee, Hoseong Seo, Dabin Kim et al.

Existing studies for environment interaction with an aerial robot have been focused on interaction with static surroundings. However, to fully explore the concept of an aerial manipulation, interaction with moving structures should also be considered. In this paper, a multirotor-based aerial manipulator opening a daily-life moving structure, a hinged door, is presented. In order to address the constrained motion of the structure and to avoid collisions during operation, model predictive control (MPC) is applied to the derived coupled system dynamics between the aerial manipulator and the door involving state constraints. By implementing a constrained version of differential dynamic programming (DDP), MPC can generate position setpoints to the disturbance observer (DOB)-based robust controller in real-time, which is validated by our experimental results.

Seung Jae Lee, Inkyu Jang, H. Jin Kim

In this paper, we introduce a new quadcopter fail-safe flight solution that can perform the same four controllable degrees-of-freedom flight as a regular multirotor even when a single thruster fails. The new solution employs a novel multirotor platform known as the T3-Multirotor and utilizes a distinctive strategy of actively controlling the center of gravity position to restore the nominal flight performance. A dedicated control structure is introduced, along with a detailed analysis of the dynamic characteristics of the platform that change during emergency flights. Experimental results are provided to validate the feasibility of the proposed fail-safe flight strategy.

Jungwon Park, H. Jin Kim

This paper presents a new trajectory planning method for multiple quadrotors in obstacle-dense environments. We suggest a relative safe flight corridor (RSFC) to model safe region between a pair of agents, and it is used to generate linear constraints for inter-collision avoidance by utilizing the convex hull property of relative Bernstein polynomial. Our approach employs a graph-based multi-agent pathfinding algorithm to generate an initial trajectory, which is used to construct a safe flight corridor (SFC) and RSFC. We express the trajectory as a piecewise Bernstein polynomial and formulate the trajectory planning problem into one quadratic programming problem using linear constraints from SFC and RSFC. The proposed method can compute collision-free trajectory for 16 agents within a second and for 64 agents less than a minute, and it is validated both through simulation and indoor flight test.