All papers

The paper presents DR-VLA, transferring outdoor navigation behaviors from egocentric human web videos to quadruped delivery robots via domain-invariant scene representations.

The paper introduces CrossFly, a simulation benchmark evaluating aerial robot navigation and subsequent physical interaction across combined indoor and outdoor environments.

The paper proposes a residual stiffening layer that dynamically adjusts robotic impedance control by detecting resistance without explicit force sensing.

The paper introduces a real-time surgical tracking system that integrates object detection, particle filtering, and monocular depth estimation to handle tool occlusions.

The paper proposes an optimization method that jointly designs robot hand mechanical parameters and retargeting algorithms to match human grasping postures faithfully.

The paper introduces a sonar-only mapping system that combines beamforming place descriptors, occupancy grid mapping, and pose graph optimization for indoor navigation.

The paper presents an attribute reliability monitor that tracks output consistency over time to detect failures in vision-language models for Earth observation.

The paper introduces WAVE-0, connecting vision-language-action policies with a lightweight latent world model to guide action generation through predicted future outcomes.

The paper introduces ForkCast, which evaluates counterfactual future outcomes using a world model to intervene and recover frozen vision-language-action manipulation policies.

The paper introduces RoboFQE, an off-policy evaluation method to assess visual manipulation policies directly from offline trajectories without generating new rollouts.

The paper introduces GreenWAM, a robotics foundation world model trained across twenty robot bodies using a shared anatomical action space.

The paper presents Equilibrium Energy Control, using an energy landscape to detect policy uncertainty and trigger expert takeovers for lifelong robot learning.

The paper introduces EgoMatrix, a pipeline that converts monocular egocentric human videos into physically executable bimanual dexterous robot demonstration trajectories.

The paper presents a method to transfer contact-rich manipulation behaviors to robots from a single demonstration using ordered contact trajectory prompts.

The paper proposes a masked diffusion framework to plan coordinated paths and recharge schedules for collaborative aerial and ground robot teams under constraints.

The paper decouples contact-rich manipulation learning into a demonstration-trained future prediction model and an inverse dynamics model trained on play data.

The paper introduces interaction latents that combine motion, contact, and force trajectories into unified action representations for contact-rich robot manipulation.

The paper introduces an agentic runtime that uses separate perception tools, rule-based verification, and feedback-driven recovery to improve closed-loop driving safety.

The paper presents EmbodiedEvo, a framework where coding agents self-improve physical manipulation policies by evolving reusable skills and lessons through execution feedback.

Orchestrates persistent multi-robot missions with language models by offloading ongoing tasks to classical protocols.

Presents a distributed semantic exploration method for robot teams to balance map information gain with movement limits.

The paper presents a decentralized multi-agent reinforcement learning controller for cooperative aerial manipulation using multiple drones.

The paper introduces an architecture that separates safe and unsafe visual knowledge during pretraining so unsafe branches can be disabled.

The paper proposes an active robot visual learning framework driven by self-supervised curiosity and world model prediction errors.

The paper presents a continuous-time neural ODE solver for simulating cloth dynamics at arbitrary timestamps.

The paper introduces a chunk-wise model-based reinforcement learning approach that uses single-pass rollouts to optimize policies without recurrent backpropagation through time.

Applies classifier-free guidance to diffusion policies to separate reinforcement learning exploration from core task capabilities.

This paper introduces learned communication adapters and action heads to coordinate multiple vision-language-action robotic policies during collaborative tasks.

The paper proposes using vision-language models as success detectors rather than reward regressors for embodied reinforcement learning.

The paper proposes a robot control framework using predictive architectures and flow matching to generate action chunks without autoregressive decoding.

This paper proposes a hierarchical control framework combining model predictive control and reinforcement learning to stabilize autonomous mobile robots with swiveling casters.

The paper introduces a trust-calibrated reinforcement learning framework coupling vision-language-action policy optimization with an evolving simulator.

The dissertation presents offline-trained neural implicit representations of reachability corrections for real-time safe robot navigation in dynamic environments.

The paper proposes an implicit integration approach for flow-based reinforcement learning policies to enable stable off-policy optimization.

The paper introduces an advantage-guided policy optimization framework that decouples policy representation from reinforcement learning policy updates.

The paper transfers quadrupedal locomotion skills from single side-view videos to robots by pinning unobserved lateral motion to stance width.

The paper proposes a multi-agent diffusion transformer framework to generate realistic air traffic scenes and trajectory patterns from recorded data.

This paper develops a deferred-decision trajectory optimization controller with conformal prediction guarantees for multi-agent target interception.

The authors propose an action quantization method that preserves value functions by minimizing reachability error for search-based robot policy learning.

The paper proposes a hierarchical reinforcement learning framework to automate multi-phase spacecraft rendezvous and docking trajectories.

The paper introduces a self-distillation framework where a vision-language model teaches itself language-conditioned robotic navigation policies without human annotations.

This paper introduces a robot manipulation benchmark to isolate failures across scaling subgoal horizons and orchestration methods.

The paper proposes a continuous probabilistic Gaussian mixture framework to improve human pose estimation under dense occlusions without retraining models.

The paper introduces a benchmark to show that visual scene semantics alone can cause multimodal models to bypass safety guardrails without adversarial prompts.

The paper introduces a multi-stage framework that corrects learned priors using X-ray measurements for sparse-view computed tomography reconstruction.

The paper presents a temporal diffusion bridge model trained on physics-based synthetic data to extract coronary arteries from angiograms without manual labels.

The paper proposes a neural block-wise auto-exposure method for single-shot high dynamic range imaging to reduce visual artifacts.

The paper examines how retinal foundation model embeddings represent macular degeneration severity and how that latent structure affects downstream prediction.

The paper audits frozen site-probe evaluation for multi-site brain MRI harmonization and finds probe accuracy is an unreliable standalone metric.

This paper proposes a multi-instance learning model with risk-aligned regularization for patient-level ultrasound diagnostic classification.

This paper introduces a training-free method adjusting token-wise guidance scales based on velocity trajectory directional changes in generative models.

This paper introduces a vision-language-action policy using hierarchical self-refinement across internal representations and action interfaces.

This paper introduces a video benchmark derived from competitive game records to evaluate dynamic spatiotemporal world state tracking.

This paper proposes a latent dynamics model to learn dense robotic manipulation rewards from videos without confounding visual distractors.

This paper presents a semantic edge-to-cloud communication method to reduce transmission bandwidth for real-time urban traffic digital twins.

This paper presents a discrete generative transformer framework for garment state estimation and dynamics prediction in robotic manipulation.

This paper evaluates category name matching versus pairwise geometry metrics for candidate retrieval in context-constrained 3D vision-language models.

This paper evaluates calibration differences between object-level softmax and agreement metrics when resolving ambiguous 3D referring queries.

This paper introduces a benchmark and a trajectory regularization method for continuous, fine-grained control in instruction-based image editing.

This paper adapts image diffusion transformers to generate 3D human motion from text using joint-time representations.