arXiv subjects

Florian Shkurti

Publications and source records attributed to Florian Shkurti.

At least 19 recordsLinked to original sources

arXiv2026

SC3-Eval: Evaluating Robot Foundation Models via Self-Consistent Video Generation

Evaluating generalist robot manipulation policies in the real world is expensive, slow, and difficult to scale. Action-conditioned video world models offer a scalable alternative by simulating policy rollouts. Autoregressive rollouts accumulate compounding errors, observations across multiple camera views must remain mutually consistent, and the evaluator must generalize to policies whose behaviors lie outside the training distribution. We address these challenges with SC3-Eval, a self-consistent video generation recipe that adapts a pre-trained video foundation model into an accurate policy evaluator by enforcing three complementary forms of consistency. First, forward-inverse dynamics consistency jointly trains the model to predict frames from actions and to recover actions from frames, anchoring generated rollouts to a physically plausible action manifold and counteracting the drift a forward-only model cannot penalize. Second, cross-view consistency trains the model to inpaint each camera view from the other, keeping the multi-camera observation coherent over long rollouts without any explicit memory mechanism. Third, test-time consistency reuses the inverse dynamics mode at inference as a per-action-chunk uncertainty signal that terminates rollouts whose generated frames drift away from the requested actions. We also demonstrate SC3-Eval rollouts reproduce the failure modes that policies exhibit in real-world rollouts, supporting fine-grained diagnostic comparison rather than aggregate ranking alone. Across seven real-world vision-language-action policies, SC3-Eval attains a closed-loop Pearson correlation of $0.929$ and MMRV of $0.119$, outperforming three strong prior video-model-based baselines, and generalizes to new tasks.

cs.RO↗

arXiv2026

Dynamic Execution Horizon Prediction for Chunk-based Robot Policies

Action chunking has become a standard design in modern robot policies, from diffusion/flow policies to vision-language-action models, where the policy predicts a sequence of actions and executes a fixed number of them instead of acting one step at a time. However, this paradigm relies on a key assumption: a fixed execution horizon. During chunk execution, the policy operates open-loop, which is particularly problematic for fine-grained manipulation tasks that require frequent replanning. In practice, the execution horizon is typically chosen through empirical tuning and is highly task-dependent. To this end, we propose Dynamic Execution Horizon Prediction (DEHP), an effective method that trains a lightweight execution-horizon prediction branch using online reinforcement learning while keeping the pretrained chunk policy completely frozen. This makes the method compatible with black-box chunk policies and isolates the effect of adapting the execution horizon from changes to the underlying action generator. Across our evaluations, DEHP improves the success rate of different high-precision and long-horizon manipulation tasks by a large margin. Our qualitative analysis further shows that DEHP predicts shorter execution horizons during fine-grained stages of the task and longer horizons during free-space motion. In this way, DEHP balances the efficiency of open-loop chunk execution with the reactivity of closed-loop single-step control. Project page: https://dehp-chunking.github.io/

cs.RO↗

arXiv2026

TAM: Torque Adaptation Module for Robust Motion Transfer in Manipulation

A policy tuned for one robot often behaves differently on another, whether due to the sim-to-real gap, unknown payloads, or the differing dynamics of two instances of the same robot. In contact-rich, dynamic manipulation, even small motion discrepancies can result in failure to track reference motion, since they disrupt the timing and modes of contact. Common remedies, such as domain randomization or system identification, either produce overly conservative task policies or require data that must be recollected for each robot or payload. We introduce the Torque Adaptation Module (TAM), a learned module that adapts the torque commands sent to the robot to match the behavior of an ideal robot. TAM operates between the low-level controller that tracks the policy's actions and the robot's torque interface. It includes a history encoder that embeds proprioceptive history into a latent state and a torque adaptor that computes residual torque corrections. Because TAM depends only on proprioceptive history and not on policy observations, or the action space, the same TAM weights can be reused to adapt policies with different action spaces (joint targets, end-effector targets, or direct torques). The policies themselves do not need to be trained with domain randomization of robot parameters. Instead, we offload the need for domain randomization to TAM by training it entirely in randomized simulation, using multi-robot pretraining followed by a robot-specific fine-tuning step that still requires no real-robot data. We evaluate TAM zero-shot on a real Franka Panda robot across dynamic manipulation tasks that include a vision-based box pushing policy (from RL), a flip policy (from BC), and an MPC ball-on-plate balancing. Our experiments show that TAM improves zero-shot real-robot execution compared to online system identification and RMA baselines and enables robust dynamic manipulation performance.

cs.RO↗

arXiv2026

Continual Model-Based Reinforcement Learning with Hypernetworks

Effective planning in model-based reinforcement learning (MBRL) and model-predictive control (MPC) relies on the accuracy of the learned dynamics model. In many instances of MBRL and MPC, this model is assumed to be stationary and is periodically re-trained from scratch on state transition experience collected from the beginning of environment interactions. This implies that the time required to train the dynamics model - and the pause required between plan executions - grows linearly with the size of the collected experience. We argue that this is too slow for lifelong robot learning and propose HyperCRL, a method that continually learns the encountered dynamics in a sequence of tasks using task-conditional hypernetworks. Our method has three main attributes: first, it includes dynamics learning sessions that do not revisit training data from previous tasks, so it only needs to store the most recent fixed-size portion of the state transition experience; second, it uses fixed-capacity hypernetworks to represent non-stationary and task-aware dynamics; third, it outperforms existing continual learning alternatives that rely on fixed-capacity networks, and does competitively with baselines that remember an ever increasing coreset of past experience. We show that HyperCRL is effective in continual model-based reinforcement learning in robot locomotion and manipulation scenarios, such as tasks involving pushing and door opening. Our project website with videos is at this link https://rvl.cs.toronto.edu/blog/hypercrl

cs.LG↗

arXiv2026

E$^3$C: Video Generation with 3D Environmental Memory and Ego-Exo Human Pose Control

Controllable and physically grounded egocentric video generation is essential for embodied agents to reason about how their own and others' actions manifest and change the world. Compared to generic video synthesis, egocentric generation is especially challenging: the camera is tightly coupled to the actor, leading to rapid viewpoint changes and frequent self-occlusions; the underlying actions are subtle, articulated, and often only partially visible; and both the people and the scene state must evolve consistently with the specified controls. We present E$^3$C, a controllable video diffusion framework for egocentric generation that builds structured and compact conditions disentangling persistent scene structure from human-driven dynamics. From context frames, E$^3$C constructs a semi-dense point cloud-based 3D memory and augments each point with appearance descriptors from video-VAE features. Rendering this memory into target viewpoints produces conditioning aligned with the target frames. Human dynamics are modeled separately. The observed people in the scene are controlled by skeleton renderings (exo human control), while the camera wearer is specified by their 3D body joints and 6DoF wrist motion (ego human control). To preserve ego human control when the wearer's body parts are invisible, we introduce an ego motion encoder that produces persistent cross-attention tokens. Experiments on Nymeria show that E$^3$C improves visual fidelity, camera-motion accuracy, object consistency, and ego & exo human control over strong baselines, while also enabling intuitive scene editing.

cs.CV↗

arXiv2026

Diversity You Can Actually Measure: A Fast, Model-Free Diversity Metric for Robotics Datasets

Robotics datasets for imitation learning typically consist of long-horizon trajectories of different lengths over states, actions, and high-dimensional observations (e.g., RGB video), making it non-trivial to quantify diversity in a way that respects the underlying trajectory structure and geometry. We extend Shannon and von Neumann entropy to this setting by defining signature transform-based entropy on the Gram matrix of a signature kernel over demonstrations, yielding entropy and diversity metrics that operate directly on the demonstration dataset. Building on these metrics, we study how dataset diversity affects generalization performance in robot imitation learning and propose a simple, model-free way to curate diverse demonstrations. We introduce FAKTUAL (FAst trajectory Kernel enTropy cUration for imitation Learning), a data curation algorithm that selects a subset of demonstrations maximizing entropy given a subset-size budget. FAKTUAL is fully model-free, requires no access to the imitation policy or rollouts, and adds negligible overhead relative to policy training. We evaluate our approach on image and state-based RoboMimic and MetaWorld benchmarks, as well as four real-world manipulation tasks. Across tasks and architectures, diversity-aware curation with FAKTUAL consistently improves downstream success rates over random selection, while being substantially more computationally efficient compared to recent robot data curation methods. Our results suggest that the entropy of demonstration datasets is a practical tool for understanding and improving dataset diversity in robot imitation learning.

cs.RO↗

arXiv2026

SMAC: Score-Matched Actor-Critics for Robust Offline-to-Online Transfer

Modern offline Reinforcement Learning (RL) methods find performant actor-critics, however, fine-tuning these actor-critics online with value-based RL algorithms typically causes immediate drops in performance. We provide evidence consistent with the hypothesis that, in the loss landscape, offline maxima for prior algorithms and online maxima are separated by low-performance valleys that gradient-based fine-tuning traverses. Following this, we present Score Matched Actor-Critic (SMAC), an offline RL method designed to learn actor-critics that transition to online value-based RL algorithms with no drop in performance. SMAC avoids valleys between offline and online maxima by regularizing the Q-function during the offline phase to respect a first-order derivative equality between the score of the policy and action-gradient of the Q-function. We experimentally demonstrate that SMAC converges to offline maxima that are connected to better online maxima via paths with monotonically increasing reward found by first-order optimization. SMAC achieves smooth transfer to Soft Actor-Critic and TD3 in 6/6 D4RL tasks. In 4/6 environments, it reduces regret by 34-58% over the best baseline.

cs.LG↗

arXiv2026

The Impact of Class Uncertainty Propagation in Perception-Based Motion Planning

Autonomous vehicles (AVs) are being increasingly deployed in urban environments. In order to operate safely and reliably, AVs need to account for the inherent uncertainty associated with perceiving the world through sensor data and incorporate that into their decision-making process. Uncertainty-aware planners have recently been developed to account for upstream perception and prediction uncertainty. However, such planners may be sensitive to prediction uncertainty miscalibration, the magnitude of which has not yet been characterized. Towards this end, we perform a detailed analysis on the impact that perceptual uncertainty propagation and calibration has on perception-based motion planning. We do so by comparing two novel prediction-planning pipelines with varying levels of uncertainty propagation on the recently-released nuPlan planning benchmark. We study the impact of upstream uncertainty calibration using closed-loop evaluation on the nuPlan challenge scenarios. We find that the method incorporating upstream uncertainty propagation demonstrates superior generalization to complex closed-loop scenarios.

cs.RO↗

arXiv2026

What Do You Need for Compositional Generalization in Diffusion Planning?

In policy learning, stitching and compositional generalization refer to the extent to which the policy is able to piece together sub-trajectories of data it is trained on to generate new and diverse behaviours. While stitching has been identified as a significant strength of offline reinforcement learning, recent generative behavioural cloning (BC) methods have also shown proficiency at stitching. However, the main factors behind this are poorly understood, hindering the development of new algorithms that can reliably stitch by design. Focusing on diffusion planners trained via generative behavioural cloning, and without resorting to dynamic programming or TD-learning, we find three properties are key enablers for composition: shift equivariance, local receptive fields, and inference choices. We use these properties to explain architecture, data, and inference choices in existing generative BC methods based on diffusion planning including replanning frequency, data augmentation, and data scaling. Our experiments show that while local receptive fields are more important than shift equivariance in creating a diffusion planner capable of composition, both are crucial. Using findings from our experiments, we develop a new architecture for diffusion planners called Eq-Net, that is simple, produces diverse trajectories competitive with more computationally expensive methods such as replanning or scaling data, and can be guided to enable generalization in goal-conditioned settings. We show that Eq-Net exhibits significant compositional generalization in a variety of navigation and manipulation tasks designed to test planning diversity.

cs.LG↗

arXiv2026

MATTERIX: toward a digital twin for robotics-assisted chemistry laboratory automation

Accelerated materials discovery is critical for addressing global challenges. However, developing new laboratory workflows relies heavily on real-world experimental trials, and this can hinder scalability because of the need for numerous physical make-and-test iterations. Here we present MATTERIX, a multiscale, graphics processing unit-accelerated robotic simulation framework designed to create high-fidelity digital twins of chemistry laboratories, thus accelerating workflow development. This multiscale digital twin simulates robotic physical manipulation, powder and liquid dynamics, device functionalities, heat transfer and basic chemical reaction kinetics. This is enabled by integrating realistic physics simulation and photorealistic rendering with a modular graphics processing unit-accelerated semantics engine, which models logical states and continuous behaviors to simulate chemistry workflows across different levels of abstraction. MATTERIX streamlines the creation of digital twin environments through open-source asset libraries and interfaces, while enabling flexible workflow design via hierarchical plan definition and a modular skill library that incorporates learning-based methods. Our approach demonstrates sim-to-real transfer in robotic chemistry setups, reducing reliance on costly real-world experiments and enabling the testing of hypothetical automated workflows in silico. The project website is available at https://accelerationconsortium.github.io/Matterix/ .

cs.RO↗

arXiv2025

Informing Acquisition Functions via Foundation Models for Molecular Discovery

Bayesian Optimization (BO) is a key methodology for accelerating molecular discovery by estimating the mapping from molecules to their properties while seeking the optimal candidate. Typically, BO iteratively updates a probabilistic surrogate model of this mapping and optimizes acquisition functions derived from the model to guide molecule selection. However, its performance is limited in low-data regimes with insufficient prior knowledge and vast candidate spaces. Large language models (LLMs) and chemistry foundation models offer rich priors to enhance BO, but high-dimensional features, costly in-context learning, and the computational burden of deep Bayesian surrogates hinder their full utilization. To address these challenges, we propose a likelihood-free BO method that bypasses explicit surrogate modeling and directly leverages priors from general LLMs and chemistry-specific foundation models to inform acquisition functions. Our method also learns a tree-structured partition of the molecular search space with local acquisition functions, enabling efficient candidate selection via Monte Carlo Tree Search. By further incorporating coarse-grained LLM-based clustering, it substantially improves scalability to large candidate sets by restricting acquisition function evaluations to clusters with statistically higher property values. We show through extensive experiments and ablations that the proposed method substantially improves scalability, robustness, and sample efficiency in LLM-guided BO for molecular discovery.

cs.LG↗

arXiv2025

Scalable Policy Evaluation with Video World Models

Training generalist policies for robotic manipulation has shown great promise, as they enable language-conditioned, multi-task behaviors across diverse scenarios. However, evaluating these policies remains difficult because real-world testing is expensive, time-consuming, and labor-intensive. It also requires frequent environment resets and carries safety risks when deploying unproven policies on physical robots. Manually creating and populating simulation environments with assets for robotic manipulation has not addressed these issues, primarily due to the significant engineering effort required and the substantial sim-to-real gap, both in terms of physics and rendering. In this paper, we explore the use of action-conditional video generation models as a scalable way to learn world models for policy evaluation. We demonstrate how to incorporate action conditioning into existing pre-trained video generation models. This allows leveraging internet-scale in-the-wild online videos during the pre-training stage and alleviates the need for a large dataset of paired video-action data, which is expensive to collect for robotic manipulation. Our paper examines the effect of dataset diversity, pre-trained weights, and common failure cases for the proposed evaluation pipeline. Our experiments demonstrate that across various metrics, including policy ranking and the correlation between actual policy values and predicted policy values, these models offer a promising approach for evaluating policies without requiring real-world interactions.

cs.RO↗

arXiv2025

Search-TTA: A Multimodal Test-Time Adaptation Framework for Visual Search in the Wild

To perform outdoor visual navigation and search, a robot may leverage satellite imagery to generate visual priors. This can help inform high-level search strategies, even when such images lack sufficient resolution for target recognition. However, many existing informative path planning or search-based approaches either assume no prior information, or use priors without accounting for how they were obtained. Recent work instead utilizes large Vision Language Models (VLMs) for generalizable priors, but their outputs can be inaccurate due to hallucination, leading to inefficient search. To address these challenges, we introduce Search-TTA, a multimodal test-time adaptation framework with a flexible plug-and-play interface compatible with various input modalities (e.g., image, text, sound) and planning methods (e.g., RL-based). First, we pretrain a satellite image encoder to align with CLIP's visual encoder to output probability distributions of target presence used for visual search. Second, our TTA framework dynamically refines CLIP's predictions during search using uncertainty-weighted gradient updates inspired by Spatial Poisson Point Processes. To train and evaluate Search-TTA, we curate AVS-Bench, a visual search dataset based on internet-scale ecological data containing 380k images and taxonomy data. We find that Search-TTA improves planner performance by up to 30.0%, particularly in cases with poor initial CLIP predictions due to domain mismatch and limited training data. It also performs comparably with significantly larger VLMs, and achieves zero-shot generalization via emergent alignment to unseen modalities. Finally, we deploy Search-TTA on a real UAV via hardware-in-the-loop testing, by simulating its operation within a large-scale simulation that provides onboard sensing.

cs.RO↗

arXiv2025

SAFE: Multitask Failure Detection for Vision-Language-Action Models

While vision-language-action models (VLAs) have shown promising robotic behaviors across a diverse set of manipulation tasks, they achieve limited success rates when deployed on novel tasks out of the box. To allow these policies to safely interact with their environments, we need a failure detector that gives a timely alert such that the robot can stop, backtrack, or ask for help. However, existing failure detectors are trained and tested only on one or a few specific tasks, while generalist VLAs require the detector to generalize and detect failures also in unseen tasks and novel environments. In this paper, we introduce the multitask failure detection problem and propose SAFE, a failure detector for generalist robot policies such as VLAs. We analyze the VLA feature space and find that VLAs have sufficient high-level knowledge about task success and failure, which is generic across different tasks. Based on this insight, we design SAFE to learn from VLA internal features and predict a single scalar indicating the likelihood of task failure. SAFE is trained on both successful and failed rollouts and is evaluated on unseen tasks. SAFE is compatible with different policy architectures. We test it on OpenVLA, $π_0$, and $π_0$-FAST in both simulated and real-world environments extensively. We compare SAFE with diverse baselines and show that SAFE achieves state-of-the-art failure detection performance and the best trade-off between accuracy and detection time using conformal prediction. More qualitative results and code can be found at the project webpage: https://vla-safe.github.io/

cs.RO↗

arXiv2025

AnyPlace: Learning Generalized Object Placement for Robot Manipulation

Object placement in robotic tasks is inherently challenging due to the diversity of object geometries and placement configurations. To address this, we propose AnyPlace, a two-stage method trained entirely on synthetic data, capable of predicting a wide range of feasible placement poses for real-world tasks. Our key insight is that by leveraging a Vision-Language Model (VLM) to identify rough placement locations, we focus only on the relevant regions for local placement, which enables us to train the low-level placement-pose-prediction model to capture diverse placements efficiently. For training, we generate a fully synthetic dataset of randomly generated objects in different placement configurations (insertion, stacking, hanging) and train local placement-prediction models. We conduct extensive evaluations in simulation, demonstrating that our method outperforms baselines in terms of success rate, coverage of possible placement modes, and precision. In real-world experiments, we show how our approach directly transfers models trained purely on synthetic data to the real world, where it successfully performs placements in scenarios where other models struggle -- such as with varying object geometries, diverse placement modes, and achieving high precision for fine placement. More at: https://any-place.github.io.

cs.RO↗

arXiv2025

Automated Planning Domain Inference for Task and Motion Planning

Task and motion planning (TAMP) frameworks address long and complex planning problems by integrating high-level task planners with low-level motion planners. However, existing TAMP methods rely heavily on the manual design of planning domains that specify the preconditions and postconditions of all high-level actions. This paper proposes a method to automate planning domain inference from a handful of test-time trajectory demonstrations, reducing the reliance on human design. Our approach incorporates a deep learning-based estimator that predicts the appropriate components of a domain for a new task and a search algorithm that refines this prediction, reducing the size and ensuring the utility of the inferred domain. Our method is able to generate new domains from minimal demonstrations at test time, enabling robots to handle complex tasks more efficiently. We demonstrate that our approach outperforms behavior cloning baselines, which directly imitate planner behavior, in terms of planning performance and generalization across a variety of tasks. Additionally, our method reduces computational costs and data amount requirements at test time for inferring new planning domains.

cs.RO↗

arXiv2025

SICNav-Diffusion: Safe and Interactive Crowd Navigation with Diffusion Trajectory Predictions

To navigate crowds without collisions, robots must interact with humans by forecasting their future motion and reacting accordingly. While learning-based prediction models have shown success in generating likely human trajectory predictions, integrating these stochastic models into a robot controller presents several challenges. The controller needs to account for interactive coupling between planned robot motion and human predictions while ensuring both predictions and robot actions are safe (i.e. collision-free). To address these challenges, we present a receding horizon crowd navigation method for single-robot multi-human environments. We first propose a diffusion model to generate joint trajectory predictions for all humans in the scene. We then incorporate these multi-modal predictions into a SICNav Bilevel MPC problem that simultaneously solves for a robot plan (upper-level) and acts as a safety filter to refine the predictions for non-collision (lower-level). Combining planning and prediction refinement into one bilevel problem ensures that the robot plan and human predictions are coupled. We validate the open-loop trajectory prediction performance of our diffusion model on the commonly used ETH/UCY benchmark and evaluate the closed-loop performance of our robot navigation method in simulation and extensive real-robot experiments demonstrating safe, efficient, and reactive robot motion.

cs.RO↗

arXiv2025

Deploying SICNav in the Field: Safe and Interactive Crowd Navigation using MPC and Bilevel Optimization

Safe and efficient navigation in crowded environments remains a critical challenge for robots that provide a variety of service tasks such as food delivery or autonomous wheelchair mobility. Classical robot crowd navigation methods decouple human motion prediction from robot motion planning, which neglects the closed-loop interactions between humans and robots. This lack of a model for human reactions to the robot plan (e.g. moving out of the way) can cause the robot to get stuck. Our proposed Safe and Interactive Crowd Navigation (SICNav) method is a bilevel Model Predictive Control (MPC) framework that combines prediction and planning into one optimization problem, explicitly modeling interactions among agents. In this paper, we present a systems overview of the crowd navigation platform we use to deploy SICNav in previously unseen indoor and outdoor environments. We provide a preliminary analysis of the system's operation over the course of nearly 7 km of autonomous navigation over two hours in both indoor and outdoor environments.

cs.RO↗