arXiv subjects

Oliver Kroemer

Publications and source records attributed to Oliver Kroemer.

At least 19 recordsLinked to original sources

arXiv2026

GTA-2: A Multi-VLM Framework for Synthesizing Robot Manipulation Skills via Grounded Task Axes

Robotic manipulation tasks are often decomposed into behaviors or skills. However, one often needs to predefine these behaviors for specific tasks or try to cover a wide range of tasks using generic skills. As a result, these behaviors can remain too coarse to expose the geometric, control, and scene-dependent decisions required for execution. We introduce Grounded Task Axes v2 (GTA-2), a modular multi-VLM framework that constructs executable, task-bespoke manipulation skills from reusable object-centric task-axis components. Rather than predicting actions end-to-end or composing fixed task-level primitives, GTA-2 represents each skill as semantic subtasks comprising task-relevant keypoints and axes, controller compositions, and scene-dependent parameters. Four specialized VLM agents separately decompose the task, construct an abstract task-axis skill, assign controller parameters, and ground the required visual features from RGB-D observations. This abstraction-to-grounding factorization enables zero-shot skill generation without task-specific robot demonstrations, policy training, or fine-tuning. It also keeps intermediate decisions explicit, allowing targeted human feedback to refine an incorrect stage while preserving correct components. We evaluate GTA-2 on 14 real-robot manipulation tasks against a VLA policy pi_{0.5} and two Code-as-Policies baselines using task-axis controllers or conventional robot primitives. GTA-2 achieves an average zero-shot success rate of 73.9%, exceeding the strongest baseline by 31.4 percentage points, while targeted refinement raises GTA-2's average success rate to 90.7%. Project page: https://gta2-project.github.io/

cs.RO↗

arXiv2026

Inference-time Policy Steering via Vision and Touch

Inference-time steering adapts pre-trained generative robot policies during deployment by verifying candidate actions before execution. While prior methods typically perform this verification only with visual observations, vision alone is often insufficient for contact-rich manipulation, where success depends on both global task progress and subtle local interactions such as contact force. We introduce ViTaL, a visuo-tactile inference-time steering framework that formulates multimodal guidance as a bi-level optimization problem. At the high level, visual sampling-and-verification performs long-horizon mode selection, deciding what behavior the robot should execute. At the low level, tactile-guided diffusion editing refines the selected action sequence over a shorter horizon to satisfy local contact requirements. To support outcome-based steering, ViTaL learns a visuo-tactile latent world model and employs semantically aligned visual and tactile verifiers, including a novel text-conditioned tactile reward that scores predicted tactile futures directly in latent space. Across three real-world contact-rich manipulation tasks, ViTaL improves overall success by 51% over the base policy, outperforms unimodal steering by at least 33%, and exceeds naive multimodal fusion by at least 20%. Website: https://yilin-wu98.github.io/vital_website.

cs.RO↗

arXiv2026

TacO: Benchmarking Tactile Sensors for Object Manipulation

Vision-based learning from demonstrations has achieved remarkable success in enabling robots to perform manipulation tasks and high-level semantic reasoning, yet it remains insufficient for complex, contact-rich manipulation. While there is broad agreement that tactile sensing improves manipulation, there is no empirical guidance on which tactile sensors are best suited for which manipulation tasks. In this paper, we provide a systematic, task-driven evaluation of tactile sensors for robot manipulation and propose a framework for selecting and evaluating sensors based on manipulation policy performance. Separate manipulation policies are trained for tactile sensors of four distinct modalities: visual, acoustic, magnetic, and resistive, across three tasks: pick-and-place with unknown mass, object reorientation, and plug insertion. For each task, an analysis of how sensor properties such as spatial resolution, shear sensing, and tactile representation, and the inherent material friction affect task performances is done. Rather than tactile sensing being universally beneficial in the same way, our results show that the usefulness of tactile information depends strongly on sensor modality, material properties, and the specific manipulation tasks. All of the tactile sensors, code, data, and hardware setup will be publicly available on the project website.

cs.RO↗

arXiv2026

OGPO: Sample Efficient Full-Finetuning of Generative Control Policies

Generative control policies (GCPs), such as diffusion- and flow-based control policies, have emerged as effective parameterizations for robot learning. This work introduces Off-policy Generative Policy Optimization (OGPO), a sample-efficient algorithm for finetuning GCPs that maintains off-policy critic networks to maximize data reuse and propagate policy gradients through the full generative process of the policy via a modified PPO objective, using critics as the terminal reward. OGPO achieves state-of-the-art performance on manipulation tasks spanning multi-task settings, high-precision insertion, and dexterous control. To our knowledge, it is also the only method that can fine-tune poorly-initialized behavior cloning policies to near full task-success with no expert data in the online replay buffer, and does so with few task-specific hyperparameter tuning. Through extensive empirical investigations, we demonstrate that OGPO drastically outperforms methods alternatives on policy steering and learning residual corrections, and identify the key mechanisms behind its performance. We further introduce practical stabilization tricks, including success-buffer regularization, two-sided conservative advantages, and Q-variance reduction, to mitigate critic over-exploitation across state- and pixel-based settings. Beyond proposing OGPO, we conduct a systematic empirical study of GCP finetuning, identifying the stabilizing mechanisms and failure modes that govern successful off-policy full-policy improvement.

cs.LG↗

arXiv2025

Autonomously Unweaving Multiple Cables Using Visual Feedback

Many cable management tasks involve separating out the different cables and removing tangles. Automating this task is challenging because cables are deformable and can have combinations of knots and multiple interwoven segments. Prior works have focused on untying knots in one cable, which is one subtask of cable management. However, in this paper, we focus on a different subtask called multi-cable unweaving, which refers to removing the intersections among multiple interwoven cables to separate them and facilitate further manipulation. We propose a method that utilizes visual feedback to unweave a bundle of loosely entangled cables. We formulate cable unweaving as a pick-and-place problem, where the grasp position is selected from discrete nodes in a graph-based cable state representation. Our cable state representation encodes both topological and geometric information about the cables from the visual image. To predict future cable states and identify valid actions, we present a novel state transition model that takes into account the straightening and bending of cables during manipulation. Using this state transition model, we select between two high-level action primitives and calculate predicted immediate costs to optimize the lower-level actions. We experimentally demonstrate that iterating the above perception-planning-action process enables unweaving electric cables and shoelaces with an 84% success rate on average.

cs.RO↗

arXiv2025

A multi-modal tactile fingertip design for robotic hands to enhance dexterous manipulation

Tactile sensing holds great promise for enhancing manipulation precision and versatility, but its adoption in robotic hands remains limited due to high sensor costs, manufacturing and integration challenges, and difficulties in extracting expressive and reliable information from signals. In this work, we present a low-cost, easy-to-make, adaptable, and compact fingertip design for robotic hands that integrates multi-modal tactile sensors. We use strain gauge sensors to capture static forces and a contact microphone sensor to measure high-frequency vibrations during contact. These tactile sensors are integrated into a compact design with a minimal sensor footprint, and all sensors are internal to the fingertip and therefore not susceptible to direct wear and tear from interactions. From sensor characterization, we show that strain gauge sensors provide repeatable 2D planar force measurements in the 0-5 N range and the contact microphone sensor has the capability to distinguish contact material properties. We apply our design to three dexterous manipulation tasks that range from zero to full visual occlusion. Given the expressiveness and reliability of tactile sensor readings, we show that different tactile sensing modalities can be used flexibly in different stages of manipulation, solely or together with visual observations to achieve improved task performance. For instance, we can precisely count and unstack a desired number of paper cups from a stack with 100\% success rate which is hard to achieve with vision only.

cs.RO↗

arXiv2025

GraphEQA: Using 3D Semantic Scene Graphs for Real-time Embodied Question Answering

In Embodied Question Answering (EQA), agents must explore and develop a semantic understanding of an unseen environment to answer a situated question with confidence. This problem remains challenging in robotics, due to the difficulties in obtaining useful semantic representations, updating these representations online, and leveraging prior world knowledge for efficient planning and exploration. To address these limitations, we propose GraphEQA, a novel approach that utilizes real-time 3D metric-semantic scene graphs (3DSGs) and task relevant images as multi-modal memory for grounding Vision-Language Models (VLMs) to perform EQA tasks in unseen environments. We employ a hierarchical planning approach that exploits the hierarchical nature of 3DSGs for structured planning and semantics-guided exploration. We evaluate GraphEQA in simulation on two benchmark datasets, HM-EQA and OpenEQA, and demonstrate that it outperforms key baselines by completing EQA tasks with higher success rates and fewer planning steps. We further demonstrate GraphEQA in multiple real-world home and office environments.

cs.RO↗

arXiv2025

Planning-Query-Guided Model Generation for Model-Based Deformable Object Manipulation

Efficient planning in high-dimensional spaces, such as those involving deformable objects, requires computationally tractable yet sufficiently expressive dynamics models. This paper introduces a method that automatically generates task-specific, spatially adaptive dynamics models by learning which regions of the object require high-resolution modeling to achieve good task performance for a given planning query. Task performance depends on the complex interplay between the dynamics model, world dynamics, control, and task requirements. Our proposed diffusion-based model generator predicts per-region model resolutions based on start and goal pointclouds that define the planning query. To efficiently collect the data for learning this mapping, a two-stage process optimizes resolution using predictive dynamics as a prior before directly optimizing using closed-loop performance. On a tree-manipulation task, our method doubles planning speed with only a small decrease in task performance over using a full-resolution model. This approach informs a path towards using previous planning and control data to generate computationally efficient yet sufficiently expressive dynamics models for new tasks.

cs.RO↗

arXiv2025

Cascaded Diffusion Models for Neural Motion Planning

Robots in the real world need to perceive and move to goals in complex environments without collisions. Avoiding collisions is especially difficult when relying on sensor perception and when goals are among clutter. Diffusion policies and other generative models have shown strong performance in solving local planning problems, but often struggle at avoiding all of the subtle constraint violations that characterize truly challenging global motion planning problems. In this work, we propose an approach for learning global motion planning using diffusion policies, allowing the robot to generate full trajectories through complex scenes and reasoning about multiple obstacles along the path. Our approach uses cascaded hierarchical models which unify global prediction and local refinement together with online plan repair to ensure the trajectories are collision free. Our method outperforms (by ~5%) a wide variety of baselines on challenging tasks in multiple domains including navigation and manipulation.

cs.RO↗

arXiv2025

Audio-Visual Contact Classification for Tree Structures in Agriculture

Contact-rich manipulation tasks in agriculture, such as pruning and harvesting, require robots to physically interact with tree structures to maneuver through cluttered foliage. Identifying whether the robot is contacting rigid or soft materials is critical for the downstream manipulation policy to be safe, yet vision alone is often insufficient due to occlusion and limited viewpoints in this unstructured environment. To address this, we propose a multi-modal classification framework that fuses vibrotactile (audio) and visual inputs to identify the contact class: leaf, twig, trunk, or ambient. Our key insight is that contact-induced vibrations carry material-specific signals, making audio effective for detecting contact events and distinguishing material types, while visual features add complementary semantic cues that support more fine-grained classification. We collect training data using a hand-held sensor probe and demonstrate zero-shot generalization to a robot-mounted probe embodiment, achieving an F1 score of 0.82. These results underscore the potential of audio-visual learning for manipulation in unstructured, contact-rich environments.

cs.RO↗

arXiv2025

Grounded Task Axes: Zero-Shot Semantic Skill Generalization via Task-Axis Controllers and Visual Foundation Models

Transferring skills between different objects remains one of the core challenges of open-world robot manipulation. Generalization needs to take into account the high-level structural differences between distinct objects while still maintaining similar low-level interaction control. In this paper, we propose an example-based zero-shot approach to skill transfer. Rather than treating skills as atomic, we decompose skills into a prioritized list of grounded task-axis (GTA) controllers. Each GTAC defines an adaptable controller, such as a position or force controller, along an axis. Importantly, the GTACs are grounded in object key points and axes, e.g., the relative position of a screw head or the axis of its shaft. Zero-shot transfer is thus achieved by finding semantically-similar grounding features on novel target objects. We achieve this example-based grounding of the skills through the use of foundation models, such as SD-DINO, that can detect semantically similar keypoints of objects. We evaluate our framework on real-robot experiments, including screwing, pouring, and spatula scraping tasks, and demonstrate robust and versatile controller transfer for each.

cs.RO↗

arXiv2025

Optimal Interactive Learning on the Job via Facility Location Planning

Collaborative robots must continually adapt to novel tasks and user preferences without overburdening the user. While prior interactive robot learning methods aim to reduce human effort, they are typically limited to single-task scenarios and are not well-suited for sustained, multi-task collaboration. We propose COIL (Cost-Optimal Interactive Learning) -- a multi-task interaction planner that minimizes human effort across a sequence of tasks by strategically selecting among three query types (skill, preference, and help). When user preferences are known, we formulate COIL as an uncapacitated facility location (UFL) problem, which enables bounded-suboptimal planning in polynomial time using off-the-shelf approximation algorithms. We extend our formulation to handle uncertainty in user preferences by incorporating one-step belief space planning, which uses these approximation algorithms as subroutines to maintain polynomial-time performance. Simulated and physical experiments on manipulation tasks show that our framework significantly reduces the amount of work allocated to the human while maintaining successful task completion.

cs.RO↗

arXiv2025

GraphSeg: Segmented 3D Representations via Graph Edge Addition and Contraction

Robots operating in unstructured environments often require accurate and consistent object-level representations. This typically requires segmenting individual objects from the robot's surroundings. While recent large models such as Segment Anything (SAM) offer strong performance in 2D image segmentation. These advances do not translate directly to performance in the physical 3D world, where they often over-segment objects and fail to produce consistent mask correspondences across views. In this paper, we present GraphSeg, a framework for generating consistent 3D object segmentations from a sparse set of 2D images of the environment without any depth information. GraphSeg adds edges to graphs and constructs dual correspondence graphs: one from 2D pixel-level similarities and one from inferred 3D structure. We formulate segmentation as a problem of edge addition, then subsequent graph contraction, which merges multiple 2D masks into unified object-level segmentations. We can then leverage \emph{3D foundation models} to produce segmented 3D representations. GraphSeg achieves robust segmentation with significantly fewer images and greater accuracy than prior methods. We demonstrate state-of-the-art performance on tabletop scenes and show that GraphSeg enables improved performance on downstream robotic manipulation tasks. Code available at https://github.com/tomtang502/graphseg.git.

cs.RO↗

arXiv2025

RecoveryChaining: Learning Local Recovery Policies for Robust Manipulation

Model-based planners and controllers are commonly used to solve complex manipulation problems as they can efficiently optimize diverse objectives and generalize to long horizon tasks. However, they often fail during deployment due to noisy actuation, partial observability and imperfect models. To enable a robot to recover from such failures, we propose to use hierarchical reinforcement learning to learn a recovery policy. The recovery policy is triggered when a failure is detected based on sensory observations and seeks to take the robot to a state from which it can complete the task using the nominal model-based controllers. Our approach, called RecoveryChaining, uses a hybrid action space, where the model-based controllers are provided as additional \emph{nominal} options which allows the recovery policy to decide how to recover, when to switch to a nominal controller and which controller to switch to even with \emph{sparse rewards}. We evaluate our approach in three multi-step manipulation tasks with sparse rewards, where it learns significantly more robust recovery policies than those learned by baselines. We successfully transfer recovery policies learned in simulation to a physical robot to demonstrate the feasibility of sim-to-real transfer with our method.

cs.RO↗

arXiv2024

SonicBoom: Contact Localization Using Array of Microphones

In cluttered environments where visual sensors encounter heavy occlusion, such as in agricultural settings, tactile signals can provide crucial spatial information for the robot to locate rigid objects and maneuver around them. We introduce SonicBoom, a holistic hardware and learning pipeline that enables contact localization through an array of contact microphones. While conventional sound source localization methods effectively triangulate sources in air, localization through solid media with irregular geometry and structure presents challenges that are difficult to model analytically. We address this challenge through a feature engineering and learning based approach, autonomously collecting 18,000 robot interaction sound pairs to learn a mapping between acoustic signals and collision locations on the robot end effector link. By leveraging relative features between microphones, SonicBoom achieves localization errors of 0.42cm for in distribution interactions and maintains robust performance of 2.22cm error even with novel objects and contact conditions. We demonstrate the system's practical utility through haptic mapping of occluded branches in mock canopy settings, showing that acoustic based sensing can enable reliable robot navigation in visually challenging environments.

cs.RO↗

arXiv2024

Autonomous Sensor Exchange and Calibration for Cornstalk Nitrate Monitoring Robot

Interactive sensors are an important component of robotic systems but often require manual replacement due to wear and tear. Automating this process can enhance system autonomy and facilitate long-term deployment. We developed an autonomous sensor exchange and calibration system for an agriculture crop monitoring robot that inserts a nitrate sensor into cornstalks. A novel gripper and replacement mechanism, featuring a reliable funneling design, were developed to enable efficient and reliable sensor exchanges. To maintain consistent nitrate sensor measurement, an on-board sensor calibration station was integrated to provide in-field sensor cleaning and calibration. The system was deployed at the Ames Curtis Farm in June 2024, where it successfully inserted nitrate sensors with high accuracy into 30 cornstalks with a 77$\%$ success rate.

cs.RO↗

arXiv2024

Tilde: Teleoperation for Dexterous In-Hand Manipulation Learning with a DeltaHand

Dexterous robotic manipulation remains a challenging domain due to its strict demands for precision and robustness on both hardware and software. While dexterous robotic hands have demonstrated remarkable capabilities in complex tasks, efficiently learning adaptive control policies for hands still presents a significant hurdle given the high dimensionalities of hands and tasks. To bridge this gap, we propose Tilde, an imitation learning-based in-hand manipulation system on a dexterous DeltaHand. It leverages 1) a low-cost, configurable, simple-to-control, soft dexterous robotic hand, DeltaHand, 2) a user-friendly, precise, real-time teleoperation interface, TeleHand, and 3) an efficient and generalizable imitation learning approach with diffusion policies. Our proposed TeleHand has a kinematic twin design to the DeltaHand that enables precise one-to-one joint control of the DeltaHand during teleoperation. This facilitates efficient high-quality data collection of human demonstrations in the real world. To evaluate the effectiveness of our system, we demonstrate the fully autonomous closed-loop deployment of diffusion policies learned from demonstrations across seven dexterous manipulation tasks with an average 90% success rate.

cs.RO↗

arXiv2024

Task-Oriented Active Learning of Model Preconditions for Inaccurate Dynamics Models

When planning with an inaccurate dynamics model, a practical strategy is to restrict planning to regions of state-action space where the model is accurate: also known as a \textit{model precondition}. Empirical real-world trajectory data is valuable for defining data-driven model preconditions regardless of the model form (analytical, simulator, learned, etc...). However, real-world data is often expensive and dangerous to collect. In order to achieve data efficiency, this paper presents an algorithm for actively selecting trajectories to learn a model precondition for an inaccurate pre-specified dynamics model. Our proposed techniques address challenges arising from the sequential nature of trajectories, and potential benefit of prioritizing task-relevant data. The experimental analysis shows how algorithmic properties affect performance in three planning scenarios: icy gridworld, simulated plant watering, and real-world plant watering. Results demonstrate an improvement of approximately 80% after only four real-world trajectories when using our proposed techniques.

cs.RO↗