Searcharxiv⌕ Search

arXiv subjects

Qing Su

Publications and source records attributed to Qing Su.

14 recordsLinked to original sources

Towards Optimism-Pessimism Trade-off in Model-based Offline-to-Online Reinforcement Learning

Model-based offline-to-online reinforcement learning (RL) enables sample-efficient adaptation by leveraging offline pre-training for online fine-tuning. However, the distribution shifts between offline and online stages often hinder fine-tuning performance. Many existing methods encode the optimism-pessimism trade-off through a single-objective formulation. Selecting its trade-off coefficient through online fine-tuning can require costly bi-level optimization. We identify the balance established during offline training as a key challenge: in our motivating experiments, optimistic policies adapt better to novel tasks, whereas pessimistic policies perform better on the same online tasks. To address this challenge, we propose a bi-objective formulation that captures this trade-off, yielding a Pareto policy pool during offline training. This policy pool enables adaptive selection during online fine-tuning across diverse online tasks. To generate the pool, we introduce Multi-Objective Soft Actor-critIC (MOSAIC), which solves bi-objective problems and constructs diverse Pareto policies. After offline training, a contextual bandit algorithm hierarchically selects the most suitable policy for fine-tuning at each online interaction step. Empirically, our pipeline, Hierarchical Pareto Policy Pool (HiP3), achieves the strongest aggregate performance across both same and novel online tasks and maintains its advantage under stronger distribution shifts. Comprehensive ablations further clarify the roles of the policy pool, MOSAIC-based pool construction, and the online selection mechanism.

cs.LG↗

Low-Rank Adaptation for Critic Learning in Off-Policy Reinforcement Learning

Scaling critic capacity is a promising direction for improving off-policy reinforcement learning (RL). However, recent work shows that larger critics are prone to overfitting and instability in replay-based bootstrapped training. In this paper, we propose using Low-Rank Adaptation (LoRA) as a structural regularizer for critic learning from scratch. Our approach freezes randomly initialized base matrices and optimizes only the corresponding low-rank adapters, thereby constraining the learned weight changes to low-rank residuals. We evaluate our method on state-based continuous-control tasks, using SAC with SimbaV2 and BRC on DMC-Hard, and FastTD3 with its SimbaV2 variant on IsaacLab. Empirically, LoRA reduces critic loss during training and improves overall policy performance, achieving the best or competitive results on most tasks. Extensive experiments demonstrate that our low-rank updates provide a simple and effective form of structural regularization for critic learning in off-policy RL.

cs.LG↗

Toward Open-World Video Segmentation over Long Horizons

Long videos challenge open-world segmentation systems to continually discover objects and preserve their identities through disappearance and reappearance. Savvy, a zero-shot, semi-online, class-agnostic system for persistent object discovery and identity maintenance, and OGA, an evaluation suite that credits coherent part-level predictions when their granularity differs from the reference annotations. Savvy combines modular mask discovery, deferred admission based on accumulated evidence, and track consolidation to maintain an evolving object set. OGA allows multiple predictions to support one reference while preserving prediction IDs, pairing granularity-aware fidelity with diagnostics of identity bleeding and fragmented support. Across 142 ScanNet videos and 106 HM3D trajectories, evaluated using adaptively sampled prefixes capped at 1,500 source frames, Savvy outperforms DEVA+SAM and EntitySAM in VPQ_inf, STQ, and AQ under both conventional and OGA evaluation. Conventional VPQ_inf reaches 18.44 versus DEVA+SAM's 11.52 on ScanNet and 17.31 versus 6.58 on HM3D. VIPSeg comparisons reveal conventional one-to-one VPQ's sensitivity to annotation granularity. Controlled identity-severing and flicker tests show that OGA detects temporal failures even when frame-level masks remain unchanged. Together, these results demonstrate complementary advances: Savvy improves long-horizon segmentation and association, and OGA distinguishes coherent part-level support from temporal identity failures. Project website: https://github.com/QingSuML/savvy

cs.CV↗

ReForge: Refining Merged Models with Anchor-Regularized Regression

Model merging aims to combine multiple task-specific expert models into a single model without joint retraining, offering a practical alternative to multi-task learning when data access or computational budget is limited. Existing model merging methods rarely exploit strong merged models as priors for further improvement. To address this limitation, we propose ReForge, a bilevel optimization framework that formulates module-wise refinement as Bayesian linear regression with an anchor-centered prior. The inner level yields a closed-form MAP estimate from unlabeled calibration activations. The outer level uses Bayesian optimization to jointly select heterogeneous regularization strengths and assembly scales using held-out validation data. Furthermore, we develop a data-free variant of ReForge that replaces activation statistics with task-vector Grams, eliminating the need for calibration examples. Across extensive benchmarks, including up to 20-task merging in vision and 5-task merging in language, ReForge consistently outperforms all evaluated plug-and-play anchor baselines (e.g., TA, WUDI-Merging, and TSV). On 20-task ViT-B/32, ReForge improves the strongest evaluated baseline, ISO-CTS, from 77.6% to 82.8% in the data-assisted setting and to 81.5% in the data-free setting. On eight-task ViT-L/14, the data-assisted variant achieves 95.1% mean accuracy, compared with 95.8% for the individual task experts. Our source code will be released soon.

cs.LG↗

REAP: Reinforcement-Learning End-to-End Autonomous Parking with Gaussian Splatting Simulator for Real2Sim2Real Transfer

In recent years, autonomous parking has made significant advances, yet parking tasks still face challenges in extreme scenarios such as mechanical and dead-end parking slots, often resulting in failures. This is mainly due to traditional parking methods adopting a multistage approach, lacking the ability to optimize the parking problem as a whole. End-to-end methods enable joint optimization across perception and planning modules to eliminate the accumulation of errors, enhancing algorithm performance in extreme scenarios. Although several end-to-end parking methods use imitation or reinforcement learning, the former is limited by data cost and distribution coverage, while the latter suffers from inefficient exploration. To address these challenges, we propose a Reinforcement learning End-to-end Autonomous Parking method (REAP). REAP employs Soft Actor-Critic (SAC) within an asymmetric reinforcement learning framework to improve training efficiency and inference performance. To accelerate model convergence, we distill the capabilities of a rule-based planner into the end-to-end network through behavior cloning. We further introduce a soft predictive collision penalty mechanism to reduce collision rates by penalizing obstacle-approaching actions. To ensure that the trained reinforcement learning network can directly transfer to real-world scenarios, we have established a Real2Sim2Real simulator. In the Real2Sim step, we use 3D Gaussian Splatting (3DGS) to transform real-world scenes into digital scenes. In the Sim2Real step, we deploy the end-to-end model onto the vehicle to bridge the Sim2Real gap. Trained in the 3DGS simulator and deployed on physical vehicles, REAP successfully parks in various types of parking spaces, especially demonstrating the feasibility of end-to-end RL parking in extremely narrow mechanical slots.

cs.RO↗

LiloDriver: A Lifelong Learning Framework for Closed-loop Motion Planning in Long-tail Autonomous Driving Scenarios

Recent advances in autonomous driving research towards motion planners that are robust, safe, and adaptive. However, existing rule-based and data-driven planners lack adaptability to long-tail scenarios, while knowledge-driven methods offer strong reasoning but face challenges in representation, control, and real-world evaluation. To address these challenges, we present LiloDriver, a lifelong learning framework for closed-loop motion planning in long-tail autonomous driving scenarios. By integrating large language models (LLMs) with a memory-augmented planner generation system, LiloDriver continuously adapts to new scenarios without retraining. It features a four-stage architecture including perception, scene encoding, memory-based strategy refinement, and LLM-guided reasoning. Evaluated on the nuPlan benchmark, LiloDriver achieves superior performance in both common and rare driving scenarios, outperforming static rule-based and learning-based planners. Our results highlight the effectiveness of combining structured memory and LLM reasoning to enable scalable, human-like motion planning in real-world autonomous driving. Our code is available at https://github.com/Hyan-Yao/LiloDriver.

cs.RO↗

LD-MoLE: Learnable Dynamic Routing for Mixture of LoRA Experts

Recent studies have shown that combining parameter-efficient fine-tuning (PEFT) with mixture-of-experts (MoE) is an effective strategy for adapting large language models (LLMs) to the downstream tasks. However, most existing approaches rely on conventional TopK routing, which requires careful hyperparameter tuning and assigns a fixed number of experts to each token. In this work, we propose LD-MoLE, a Learnable Dynamic routing mechanism for Mixture of LoRA Experts that enables adaptive, token-dependent, and layer-wise expert allocation. Our method replaces the non-differentiable TopK selection with a differentiable routing function and a closed-form solution. Moreover, our design allows the model to adaptively determine the number of experts to activate for each token at different layers. In addition, we introduce an analytical sparsity control objective to regularize the number of activated experts. Extensive experiments on the Qwen3-1.7B and Llama-3.2-3B models show that LD-MoLE achieves the highest average scores compared to state-of-the-art baselines, across a diverse set of benchmarks. Our method not only achieves superior performance, but also demonstrates the ability to learn token-dependent and layer-wise expert allocation.

cs.CL↗

Uni-LoRA: One Vector is All You Need

Low-Rank Adaptation (LoRA) has become the de facto parameter-efficient fine-tuning (PEFT) method for large language models (LLMs) by constraining weight updates to low-rank matrices. Recent works such as Tied-LoRA, VeRA, and VB-LoRA push efficiency further by introducing additional constraints to reduce the trainable parameter space. In this paper, we show that the parameter space reduction strategies employed by these LoRA variants can be formulated within a unified framework, Uni-LoRA, where the LoRA parameter space, flattened as a high-dimensional vector space $R^D$, can be reconstructed through a projection from a subspace R^d, with $d \ll D$. We demonstrate that the fundamental difference among various LoRA methods lies in the choice of the projection matrix, $P \in R^{D \times d}$.Most existing LoRA variants rely on layer-wise or structure-specific projections that limit cross-layer parameter sharing, thereby compromising parameter efficiency. In light of this, we introduce an efficient and theoretically grounded projection matrix that is isometric, enabling global parameter sharing and reducing computation overhead. Furthermore, under the unified view of Uni-LoRA, this design requires only a single trainable vector to reconstruct LoRA parameters for the entire LLM - making Uni-LoRA both a unified framework and a "one-vector-only" solution. Extensive experiments on GLUE, mathematical reasoning, and instruction tuning benchmarks demonstrate that Uni-LoRA achieves state-of-the-art parameter efficiency while outperforming or matching prior approaches in predictive performance. Our code is available at https://github.com/KaiyangLi1992/Uni-LoRA.

cs.LG↗

Unsqueeze [CLS] Bottleneck to Learn Rich Representations

Distillation-based self-supervised learning typically leads to more compressed representations due to its radical clustering process and the implementation of a sharper target distribution. To overcome this limitation and preserve more information from input, we introduce UDI, conceptualized as Unsqueezed Distillation-based self-supervised learning (SSL). UDI enriches the learned representation by encouraging multimodal prediction distilled from a consolidated profile of local predictions that are derived via stratified sampling. Our evaluations show that UDI not only promotes semantically meaningful representations at instance level, delivering superior or competitive results to state-of-the-art SSL methods in image classification, but also effectively preserves the nuisance of input, which yields significant improvement in dense prediction tasks, including object detection and segmentation. Additionally, UDI performs competitively in low-shot image classification, improving the scalability of joint-embedding pipelines. Various visualizations and ablation studies are presented to further elucidate the mechanisms behind UDI. Our source code is available at https://github.com/ISL-CV/udi.

cs.CV↗

FLSL: Feature-level Self-supervised Learning

Current self-supervised learning (SSL) methods (e.g., SimCLR, DINO, VICReg,MOCOv3) target primarily on representations at instance level and do not generalize well to dense prediction tasks, such as object detection and segmentation.Towards aligning SSL with dense predictions, this paper demonstrates for the first time the underlying mean-shift clustering process of Vision Transformers (ViT), which aligns well with natural image semantics (e.g., a world of objects and stuffs). By employing transformer for joint embedding and clustering, we propose a two-level feature clustering SSL method, coined Feature-Level Self-supervised Learning (FLSL). We present the formal definition of the FLSL problem and construct the objectives from the mean-shift and k-means perspectives. We show that FLSL promotes remarkable semantic cluster representations and learns an embedding scheme amenable to intra-view and inter-view feature clustering. Experiments show that FLSL yields significant improvements in dense prediction tasks, achieving 44.9 (+2.8)% AP and 46.5% AP in object detection, as well as 40.8 (+2.3)% AP and 42.1% AP in instance segmentation on MS-COCO, using Mask R-CNN with ViT-S/16 and ViT-S/8 as backbone, respectively. FLSL consistently outperforms existing SSL methods across additional benchmarks, including UAV17 object detection on UAVDT, and video instance segmentation on DAVIS 2017.We conclude by presenting visualization and various ablation studies to better understand the success of FLSL. The source code is available at https://github.com/ISL-CV/FLSL.

cs.LG↗

ChiTransformer:Towards Reliable Stereo from Cues

Current stereo matching techniques are challenged by restricted searching space, occluded regions and sheer size. While single image depth estimation is spared from these challenges and can achieve satisfactory results with the extracted monocular cues, the lack of stereoscopic relationship renders the monocular prediction less reliable on its own, especially in highly dynamic or cluttered environments. To address these issues in both scenarios, we present an optic-chiasm-inspired self-supervised binocular depth estimation method, wherein a vision transformer (ViT) with gated positional cross-attention (GPCA) layers is designed to enable feature-sensitive pattern retrieval between views while retaining the extensive context information aggregated through self-attentions. Monocular cues from a single view are thereafter conditionally rectified by a blending layer with the retrieved pattern pairs. This crossover design is biologically analogous to the optic-chasma structure in the human visual system and hence the name, ChiTransformer. Our experiments show that this architecture yields substantial improvements over state-of-the-art self-supervised stereo approaches by 11%, and can be used on both rectilinear and non-rectilinear (e.g., fisheye) images. Project is available at https://github.com/ISL-CV/ChiTransformer.

cs.CV↗

Towards Bridging the Performance Gaps of Joint Energy-based Models

Can we train a hybrid discriminative-generative model within a single network? This question has recently been answered in the affirmative, introducing the field of Joint Energy-based Model (JEM), which achieves high classification accuracy and image generation quality simultaneously. Despite recent advances, there remain two performance gaps: the accuracy gap to the standard softmax classifier, and the generation quality gap to state-of-the-art generative models. In this paper, we introduce a variety of training techniques to bridge the accuracy gap and the generation quality gap of JEM. 1) We incorporate a recently proposed sharpness-aware minimization (SAM) framework to train JEM, which promotes the energy landscape smoothness and the generalizability of JEM. 2) We exclude data augmentation from the maximum likelihood estimate pipeline of JEM, and mitigate the negative impact of data augmentation to image generation quality. Extensive experiments on multiple datasets demonstrate that our SADA-JEM achieves state-of-the-art performances and outperforms JEM in image classification, image generation, calibration, out-of-distribution detection and adversarial robustness by a notable margin. Our code is available at https://github.com/sndnyang/SADAJEM.

cs.CV↗

RoadMap: A Light-Weight Semantic Map for Visual Localization towards Autonomous Driving

Accurate localization is of crucial importance for autonomous driving tasks. Nowadays, we have seen a lot of sensor-rich vehicles (e.g. Robo-taxi) driving on the street autonomously, which rely on high-accurate sensors (e.g. Lidar and RTK GPS) and high-resolution map. However, low-cost production cars cannot afford such high expenses on sensors and maps. How to reduce costs? How do sensor-rich vehicles benefit low-cost cars? In this paper, we proposed a light-weight localization solution, which relies on low-cost cameras and compact visual semantic maps. The map is easily produced and updated by sensor-rich vehicles in a crowd-sourced way. Specifically, the map consists of several semantic elements, such as lane line, crosswalk, ground sign, and stop line on the road surface. We introduce the whole framework of on-vehicle mapping, on-cloud maintenance, and user-end localization. The map data is collected and preprocessed on vehicles. Then, the crowd-sourced data is uploaded to a cloud server. The mass data from multiple vehicles are merged on the cloud so that the semantic map is updated in time. Finally, the semantic map is compressed and distributed to production cars, which use this map for localization. We validate the performance of the proposed map in real-world experiments and compare it against other algorithms. The average size of the semantic map is $36$ kb/km. We highlight that this framework is a reliable and practical localization solution for autonomous driving.

cs.CV↗

AVP-SLAM: Semantic Visual Mapping and Localization for Autonomous Vehicles in the Parking Lot

Autonomous valet parking is a specific application for autonomous vehicles. In this task, vehicles need to navigate in narrow, crowded and GPS-denied parking lots. Accurate localization ability is of great importance. Traditional visual-based methods suffer from tracking lost due to texture-less regions, repeated structures, and appearance changes. In this paper, we exploit robust semantic features to build the map and localize vehicles in parking lots. Semantic features contain guide signs, parking lines, speed bumps, etc, which typically appear in parking lots. Compared with traditional features, these semantic features are long-term stable and robust to the perspective and illumination change. We adopt four surround-view cameras to increase the perception range. Assisting by an IMU (Inertial Measurement Unit) and wheel encoders, the proposed system generates a global visual semantic map. This map is further used to localize vehicles at the centimeter level. We analyze the accuracy and recall of our system and compare it against other methods in real experiments. Furthermore, we demonstrate the practicability of the proposed system by the autonomous parking application.

cs.RO↗