arXiv subjects

Peyman Moghadam

Publications and source records attributed to Peyman Moghadam.

At least 19 recordsLinked to original sources

arXiv2026

NEO: NeRF It Once, Edit It Many Times for Continuous Object Manipulation

In this paper, we present NEO, a unified framework providing language-guided NeRF editing for robotic manipulation. Our paper introduces (i) a language-guided object removal that combines neural field resampling with multiview-consistent progressive inpainting, (ii) a direct NeRF weight editing method utilizing knowledge distillation, composing original and edited NeRFs via a teacher-student model, enabling coherent modeling of future scene states before a robot executes an action, and (iii) the first benchmark (NEO-Dataset) for quantitatively evaluating NeRF scene editing methods suitable for robot manipulation. We show that our approach outperforms state-of-the-art baselines in scene editing tasks, including object removal and pick-and-place robotic experiments, yielding visually coherent and geometrically consistent edits that reduce artifacts commonly introduced by prior methods.

cs.RO↗

arXiv2026

TOLiD: Bridging the Architecture Gap in Vision Foundation Model to LiDAR Pretraining via Token Lifting for Distillation

Cross-modal distillation from Vision Foundation Models (VFMs) to LiDAR backbones has recently emerged as a self-supervised pretraining strategy that reduces reliance on dense point-wise annotation for 3D scene understanding. However, existing distillation pipelines typically treat the VFM as a frozen feature source and train a heterogeneous 3D backbone to match fixed image embeddings, forcing the student to bridge both the modality gap and the cross-architecture gap between dense ViT token representations and sparse 3D encoders. We propose TOLiD, a self-supervised pretraining method for LiDAR representation learning that addresses this gap by coupling a LiDAR backbone with a student Vision Transformer (ViT) initialized from a frozen VFM teacher and applying supervision over compatible patch-token representations. TOLiD converts the set of point features within each image patch frustum into a token using Frustum Pooling followed by Frustum Attention, and performs token-level distillation with visibility masking. For LiDAR-only deployment, we lift token features back to per-point representations using masked bilinear sampling to avoid patches that have limited LiDAR points. We extensively evaluate TOLiD on five heterogeneous LiDAR datasets and four cross-sensor adaptation pairs, demonstrating improved transfer with frozen backbones and lightweight heads.

cs.CV↗

arXiv2026

GemNav: Discrete-Token Visual Robot Navigation using a Multimodal Large Language Model

Visual navigation policies built on large pretrained models have so far followed a common recipe: a dedicated visual encoder, a bespoke action head, and training on thousands of hours of cross-embodiment datasets. We ask whether this recipe is necessary. In this paper, we introduce GemNav, a visual robot navigation policy that adapts a frozen Multimodal Large Language Model (MLLM) for short-to-medium horizon waypoint navigation using Low-Rank Adaptation (LoRA) on the language tower alone, with no auxiliary visual encoder and no continuous regression head. Waypoints and categorical navigation signals share a single discrete token vocabulary generated by the language-model head, and a soft-decoded auxiliary loss recovers the metric structure that pure cross-entropy training discards. On a single 8.7-hour open corpus, roughly three orders of magnitude smaller than competing training sets, the policy transfers zero-shot to four physically distinct unseen environments and stops within 0.25-0.42m of the goal across 20 real-world trials covering an open carpark, an obstacle carpark, a long outdoor chemical yard, and an indoor warehouse. Conditioning on short image histories improves offline metrics but yields no robot benefit, pointing to a ceiling on what temporal context adds once pretrained vision features are in place. These results indicate that discrete-token adaptation of frozen MLLMs can provide a data-efficient, deployable alternative for foundation model robot navigation.

cs.RO↗

arXiv2026

Spatially Stratified Distillation for Heterogeneous Radar Place Recognition

Scalable, all-weather place recognition increasingly relies on heterogeneous radar place recognition to bridge diverse hardware platforms. A notable application is matching queries from cost-effective 4D automotive radars against high-fidelity reference maps built by dense spinning radars. This process is fundamentally limited by the extreme sparsity (and narrow field-of-view) of the 4D sensor, which captures only a fraction of the structural density present in the spinning radar database. Prior efforts address this issue by unifying different radar signals. That is, projecting both signals into a common representational space. Yet, they suffer performance degradation in multi-session environments. In this paper, we propose spatially-stratified distillation (SSD); a strategy that replaces standard uniform distillation with an asymmetric spatial alignment derived directly from physical radar returns. In regions where both radars exhibit overlapping returns, SSD enforces strong feature alignment. Crucially, in sparse regions where the 4D student lacks returns but the teacher contains valid structure within the shared field of view, SSD applies heavily discounted distillation weights. Extensive evaluations of the recent HeRCULES dataset demonstrate that SSD significantly outperforms prior place recognition methods, achieving state-of-the-art results on its challenging dynamic sequences.

cs.CV↗

arXiv2026

Visual Place Recognition in Forests with Depth-Aware Distillation

Visual place recognition in natural forest environments remains challenging due to repetitive vegetation, weak structural cues, and significant appearance variation across traversals. To address this limitation, this paper proposes a lightweight depth-aware distillation framework that injects geometric cues into a DINOv2-based place recognition model, while maintaining its pre-trained descriptor space. Evaluated on the recent WildCross benchmark, the proposed approach yields gains over an appearance-only counterpart, providing robustness to appearance variations. These results demonstrate the importance of depth as a strong complementary modality for place recognition in natural environments and identify depth-aware distillation as a promising direction for more robust forest perception.

cs.CV↗

arXiv2026

Cross-Modal Benchmarking for Robotic Perception in Natural Environments

Natural environments present a complex challenge to robotics perception systems. Current models, particularly vision foundation models, are largely trained on structured, urban environments leading to weaknesses in their perception for field robotics tasks. We showcase the limitations of current models using our recently released WildCross benchmark, a new cross-modal benchmark for place recognition and metric depth estimation in large-scale natural environments. WildCross comprises over 476K sequential RGB frames with semi-dense depth and surface normal annotations, each aligned with accurate 6DoF pose and synchronized dense lidar submaps. In this work, we provide an expanded analysis of the benchmark results from the recent WildCross benchmark, with particular emphasis on expanded metric depth estimation experiments. Access to the code repository and dataset for this work can be found at https://csiro-robotics.github.io/WildCross.

cs.CV↗

arXiv2026

The Quantization Benefits of Residual-Free Transformers

Large-scale transformer training and deployment are increasingly constrained by the transfer of activations, gradients, and optimizer states across accelerators. Low-bit quantization offers a natural remedy, but transformer activations are often heavy-tailed and outlier-dominated, making simple quantization highly lossy. We show that this difficulty is not only a property of the quantizer, but also of the architecture. Specifically, residual connections can drive transformer activations away from Gaussianity during training. Using controlled comparisons between residual and residual-free transformers, we demonstrate that this effect leads to substantially higher quantization error and accuracy degradation at low precision in residual models. We explain the phenomenon through an excess kurtosis analysis, showing that residual mixing can amplify non-Gaussianity, whereas dense mixing in residual-free contracts non-Gaussianity. We then show that residual-free transformers can be made trainable using orthogonal initialization, spectral or second-order optimization, and depth-aware scaling of attention temperature. In language tasks, while there is a small drop in full precision performance, these models retain near-Gaussian activations and exhibit significantly improved robustness to low-bit quantization. Our results identify an accuracy--compressibility trade-off in transformer design and motivate architecture-level approaches to quantization-friendly foundation models.

cs.LG↗

arXiv2026

Ilov3Splat: Instance-Level Open-Vocabulary 3D Scene Understanding in Gaussian Splatting

We introduce Ilov3Splat, a novel framework for instance-level open-vocabulary 3D scene understanding built on 3D Gaussian Splatting (3D-GS). Most prior work depends on 2D rendering-based matching or point-level semantic association, which undermines cross-view consistency, lacks coherent instance-level reasoning, and limits precision in downstream 3D tasks. To address these limitations, our method jointly optimizes scene geometry and semantic representations by augmenting Gaussian splats with view-consistent feature fields. Specifically, we leverage multi-resolution hash embedding to efficiently encode language-aligned CLIP features, enabling dense and coherent language grounding in 3D space. We further train an instance feature field using contrastive loss over SAM masks, supporting fine-grained object distinction across views. At inference time, CLIP-encoded queries are matched against the learned features, followed by two-stage 3D clustering to retrieve relevant Gaussian groups. This enables our framework to identify arbitrary objects in 3D scenes based on natural language descriptions, without requiring category supervision or manual annotations. Experiments on standard benchmarks demonstrate that Ilov3Splat outperforms prior open-vocabulary 3D-GS methods in both object selection and instance segmentation, offering a flexible and accurate solution for language-driven 3D scene understanding. Project page: https://csiro-robotics.github.io/Ilov3Splat.

cs.CV↗

arXiv2026

SPREAD: Subspace Representation Distillation for Lifelong Imitation Learning

A key challenge in lifelong imitation learning (LIL) is enabling agents to acquire new skills from expert demonstrations while retaining prior knowledge. This requires preserving the low-dimensional manifolds and geometric structures that underlie task representations across sequential learning. Existing distillation methods, which rely on L2-norm feature matching in raw feature space, are sensitive to noise and high-dimensional variability, often failing to preserve intrinsic task manifolds. To address this, we introduce SPREAD, a geometry-preserving framework that employs singular value decomposition (SVD) to align policy representations across tasks within low-rank subspaces. This alignment maintains the underlying geometry of multimodal features, facilitating stable transfer, robustness, and generalization. Additionally, we propose a confidence-guided distillation strategy that applies a Kullback-Leibler divergence loss restricted to the top-M most confident action samples, emphasizing reliable modes and improving optimization stability. Experiments on the LIBERO, lifelong imitation learning benchmark, show that SPREAD substantially improves knowledge transfer, mitigates catastrophic forgetting, and achieves state-of-the-art performance.

cs.LG↗

arXiv2026

WildCross: A Cross-Modal Large Scale Benchmark for Place Recognition and Metric Depth Estimation in Natural Environments

Recent years have seen a significant increase in demand for robotic solutions in unstructured natural environments, alongside growing interest in bridging 2D and 3D scene understanding. However, existing robotics datasets are predominantly captured in structured urban environments, making them inadequate for addressing the challenges posed by complex, unstructured natural settings. To address this gap, we propose WildCross, a cross-modal benchmark for place recognition and metric depth estimation in large-scale natural environments. WildCross comprises over 476K sequential RGB frames with semi-dense depth and surface normal annotations, each aligned with accurate 6DoF poses and synchronized dense lidar submaps. We conduct comprehensive experiments on visual, lidar, and cross-modal place recognition, as well as metric depth estimation, demonstrating the value of WildCross as a challenging benchmark for multi-modal robotic perception tasks. We provide access to the code repository and dataset at https://csiro-robotics.github.io/WildCross.

cs.CV↗

arXiv2025

Point-PNG: Conditional Pseudo-Negatives Generation for Point Cloud Pre-Training

We propose Point-PNG, a novel self-supervised learning framework that generates conditional pseudo-negatives in the latent space to learn point cloud representations that are both discriminative and transformation-sensitive. Conventional self-supervised learning methods focus on achieving invariance, discarding transformation-specific information. Recent approaches incorporate transformation sensitivity by explicitly modeling relationships between original and transformed inputs. However, they often suffer from an invariant-collapse phenomenon, where the predictor degenerates into identity mappings, resulting in latent representations with limited variation across transformations. To address this, we propose Point-PNG that explicitly penalizes invariant collapse through pseudo-negatives generation, enabling the network to capture richer transformation cues while preserving discriminative representations. To this end, we introduce a parametric network, COnditional Pseudo-Negatives Embedding (COPE), which learns localized displacements induced by transformations within the latent space. A key challenge arises when jointly training COPE with the MAE, as it tends to converge to trivial identity mappings. To overcome this, we design a loss function based on pseudo-negatives conditioned on the transformation, which penalizes such trivial invariant solutions and enforces meaningful representation learning. We validate Point-PNG on shape classification and relative pose estimation tasks, showing competitive performance on ModelNet40 and ScanObjectNN under challenging evaluation protocols, and achieving superior accuracy in relative pose estimation compared to supervised baselines.

cs.CV↗

arXiv2025

Cutting the Skip: Training Residual-Free Transformers

Transformers have achieved remarkable success across a wide range of applications, a feat often attributed to their scalability. Yet training them without skip (residual) connections remains notoriously difficult. While skips stabilize optimization, they also disrupt the hierarchical structure of representations, raising the long-standing question of whether transformers can be trained efficiently without them. In this work, we address this problem by analyzing the Jacobian of a skipless transformer block, showing why skips improve conditioning and revealing that their stabilization benefits can be recovered through a principled initialization strategy. Building on this insight, we introduce the first method that enables stable and efficient training of skipless transformers without altering the standard architecture. We validate our approach on Vision Transformers (ViTs) in both supervised and self-supervised settings, demonstrating that skipless ViTs trained with our initialization overcome the usual optimization barriers, learn richer hierarchical representations, and outperform strong baselines, that incorporate skip connections, on dense prediction benchmarks. These results show that skip connections are not a fundamental requirement for training ViTs and open new avenues for hierarchical representation learning in vision models.

cs.LG↗

arXiv2025

Always Skip Attention

We highlight a curious empirical result within modern Vision Transformers (ViTs). Specifically, self-attention catastrophically fails to train unless it is used in conjunction with a skip connection. This is in contrast to other elements of a ViT that continue to exhibit good performance (albeit suboptimal) when skip connections are removed. Further, we show that this critical dependence on skip connections is a relatively new phenomenon, with previous deep architectures (\eg, CNNs) exhibiting good performance in their absence. In this paper, we theoretically characterize that the self-attention mechanism is fundamentally ill-conditioned and is, therefore, uniquely dependent on skip connections for regularization. Additionally, we propose Token Graying -- a simple yet effective complement (to skip connections) that further improves the condition of input tokens. We validate our approach in both supervised and self-supervised training methods.

cs.LG↗

arXiv2025

Flashbacks to Harmonize Stability and Plasticity in Continual Learning

We introduce Flashback Learning (FL), a novel method designed to harmonize the stability and plasticity of models in Continual Learning (CL). Unlike prior approaches that primarily focus on regularizing model updates to preserve old information while learning new concepts, FL explicitly balances this trade-off through a bidirectional form of regularization. This approach effectively guides the model to swiftly incorporate new knowledge while actively retaining its old knowledge. FL operates through a two-phase training process and can be seamlessly integrated into various CL methods, including replay, parameter regularization, distillation, and dynamic architecture techniques. In designing FL, we use two distinct knowledge bases: one to enhance plasticity and another to improve stability. FL ensures a more balanced model by utilizing both knowledge bases to regularize model updates. Theoretically, we analyze how the FL mechanism enhances the stability-plasticity balance. Empirically, FL demonstrates tangible improvements over baseline methods within the same training budget. By integrating FL into at least one representative baseline from each CL category, we observed an average accuracy improvement of up to 4.91% in Class-Incremental and 3.51% in Task-Incremental settings on standard image classification benchmarks. Additionally, measurements of the stability-to-plasticity ratio confirm that FL effectively enhances this balance. FL also outperforms state-of-the-art CL methods on more challenging datasets like ImageNet.

cs.LG↗

arXiv2025

A Deeper Look into Second-Order Feature Aggregation for LiDAR Place Recognition

Efficient LiDAR Place Recognition (LPR) compresses dense pointwise features into compact global descriptors. While first-order aggregators such as GeM and NetVLAD are widely used, they overlook inter-feature correlations that second-order aggregation naturally captures. Full covariance, a common second-order aggregator, is high in dimensionality; as a result, practitioners often insert a learned projection or employ random sketches -- both of which either sacrifice information or increase parameter count. However, no prior work has systematically investigated how first- and second-order aggregation perform under constrained feature and compute budgets. In this paper, we first demonstrate that second-order aggregation retains its superiority for LPR even when channels are pruned and backbone parameters are reduced. Building on this insight, we propose Channel Partition-based Second-order Local Feature Aggregation (CPS): a drop-in, partition-based second-order aggregation module that preserves all channels while producing an order-of-magnitude smaller descriptor. CPS matches or exceeds the performance of full covariance and outperforms random projection variants, delivering new state-of-the-art results with only four additional learnable parameters across four large-scale benchmarks: Oxford RobotCar, In-house, MulRan, and WildPlaces.

cs.CV↗

arXiv2025

Shape-Space Deformer: Unified Visuo-Tactile Representations for Robotic Manipulation of Deformable Objects

Accurate modelling of object deformations is crucial for a wide range of robotic manipulation tasks, where interacting with soft or deformable objects is essential. Current methods struggle to generalise to unseen forces or adapt to new objects, limiting their utility in real-world applications. We propose Shape-Space Deformer, a unified representation for encoding a diverse range of object deformations using template augmentation to achieve robust, fine-grained reconstructions that are resilient to outliers and unwanted artefacts. Our method improves generalization to unseen forces and can rapidly adapt to novel objects, significantly outperforming existing approaches. We perform extensive experiments to test a range of force generalisation settings and evaluate our method's ability to reconstruct unseen deformations, demonstrating significant improvements in reconstruction accuracy and robustness. Our approach is suitable for real-time performance, making it ready for downstream manipulation applications.

cs.RO↗

arXiv2025

Dual-Domain Masked Image Modeling: A Self-Supervised Pretraining Strategy Using Spatial and Frequency Domain Masking for Hyperspectral Data

Hyperspectral images (HSIs) capture rich spectral signatures that reveal vital material properties, offering broad applicability across various domains. However, the scarcity of labeled HSI data limits the full potential of deep learning, especially for transformer-based architectures that require large-scale training. To address this constraint, we propose Spatial-Frequency Masked Image Modeling (SFMIM), a self-supervised pretraining strategy for hyperspectral data that utilizes the large portion of unlabeled data. Our method introduces a novel dual-domain masking mechanism that operates in both spatial and frequency domains. The input HSI cube is initially divided into non-overlapping patches along the spatial dimension, with each patch comprising the entire spectrum of its corresponding spatial location. In spatial masking, we randomly mask selected patches and train the model to reconstruct the masked inputs using the visible patches. Concurrently, in frequency masking, we remove portions of the frequency components of the input spectra and predict the missing frequencies. By learning to reconstruct these masked components, the transformer-based encoder captures higher-order spectral-spatial correlations. We evaluate our approach on three publicly available HSI classification benchmarks and demonstrate that it achieves state-of-the-art performance. Notably, our model shows rapid convergence during fine-tuning, highlighting the efficiency of our pretraining strategy.

cs.CV↗

arXiv2025

Plant species richness prediction from DESIS hyperspectral data: A comparison study on feature extraction procedures and regression models

The diversity of terrestrial vascular plants plays a key role in maintaining the stability and productivity of ecosystems. Airborne hyperspectral imaging has shown promise for measuring plant diversity remotely, but to operationalise these efforts over large regions we need to advance satellite-based alternatives. The advanced spectral and spatial specification of the recently launched DESIS (the DLR Earth Sensing Imaging Spectrometer) instrument provides a unique opportunity to test the potential for monitoring plant species diversity with spaceborne hyperspectral data. This study provides a quantitative assessment on the ability of DESIS hyperspectral data for predicting plant species richness in two different habitat types in southeast Australia. Spectral features were first extracted from the DESIS spectra, then regressed against on-ground estimates of plant species richness, with a two-fold cross validation scheme to assess the predictive performance. We tested and compared the effectiveness of Principal Component Analysis (PCA), Canonical Correlation Analysis (CCA), and Partial Least Squares analysis (PLS) for feature extraction, and Kernel Ridge Regression (KRR), Gaussian Process Regression (GPR), and Random Forest Regression (RFR) for species richness prediction. The best prediction results were $r=0.76$ and $\text{RMSE}=5.89$ for the Southern Tablelands region, and $r=0.68$ and $\text{RMSE}=5.95$ for the Snowy Mountains region. Relative importance analysis for the DESIS spectral bands showed that the red-edge, red, and blue spectral regions were more important for predicting plant species richness than the green bands and the near-infrared bands beyond red-edge. We also found that the DESIS hyperspectral data performed better than Sentinel-2 multispectral data in the prediction of plant species richness.

cs.LG↗