All papers

The paper introduces a visual self-modeling method that learns explicit screw-axis robot kinematics directly from multi-view images and joint readings.

The paper develops an end-to-end vision model that directly generates navigation trajectories for quadrotors from stereo camera images without explicit depth mapping.

The paper presents a textile sensor combining muscle vibrations and strain measurements to trigger assistance in soft knee exosuits.

The paper presents a data collection framework that aligns teacher demonstrations with robot interface actions for open-world mobile manipulation.

The paper proposes a multi-agent reinforcement learning approach using graph attention to coordinate automated vehicles at unsignalized intersections under imperfect communications.

The paper presents a streaming world action modeling framework that improves temporal continuity and execution smoothness in robot policy rollouts.

The paper introduces a supervised learning method to compose movement primitives into complex robotic manipulation tasks using unsegmented trajectories.

The paper develops a predictive control framework enabling bipedal robots to cross obstacles smoothly across both walking and running speeds.

The paper presents an algorithm for distributed multi-robot area coverage path planning that explicitly incorporates headland paths.

The paper describes a multi-sensor system using lidar, RTK positioning, and thrust feedback for mid-air drone launch and recovery.

The paper develops a linear dynamic model and control method to prevent trailer sway and jackknifing in autonomous passenger vehicles.

The paper introduces a reinforcement learning framework that optimizes the physical arm morphology of a planar manipulator to reduce real-world motor effort.

The paper presents a competitive task allocation framework using temporal vision windows and artificial potential fields for dual-arm industrial manipulation.

The paper proposes a flow-matching method initialized from recent history to speed up training and inference in robot foundation models.

The paper introduces a framework for robotic manipulation that enables self-improvement by attributing downstream failures to earlier actions without demonstrations or fine-tuning.

The paper presents a framework that factorizes humanoid control into terrain-invariant intent and perceptive terrain-conditioned realization using reinforcement learning.

This paper presents a learned onboard proprioceptive reflex policy enabling quadrotors to recover stability autonomously after high-speed collisions.

This paper proposes a signal temporal logic model predictive control method combining hard safety constraints with soft interaction quality objectives.

This paper proposes a Jacobian-augmented kinematic safety filter enforcing joint limits and task-space viability through nested quadratic programs.

This paper presents an open hardware-partitioned miniature vehicle separating onboard safety control from off-board neural model inference.

This paper presents a depth-conditioned adapter that converts unitless trajectory predictions from visual navigation policies into metric paths.

This paper develops a constrained differentiable optimization method for quadrotor obstacle avoidance using smooth clearance functions.

This paper proposes a Riemannian pullback regularization method enforcing non-holonomic vehicle kinematics into latent world models for autonomous driving.

This paper introduces a training method using reconstruction-guided 3D flow to teach world dynamics to vision-language-action models.

This paper presents a multi-floating-link parallel wire-driven robot system coupled via an inter-body wire.

This paper adapts dynamic movement primitives across segmented pouring phases for one-shot learning of latte art manipulation.

This paper introduces a benchmark evaluating how language model agents reuse embodied skills to complete physical tasks on robots.

This paper evaluates the sensitivity of robot world-action models to rephrased spatial language instructions across manipulation scenes.

This paper introduces a tactile conditioning framework separating contact signals into distinct temporal roles for robotic insertion policies.

This paper proposes a runtime failure monitor combining visual anomaly detection with action-conditioned future risk forecasting for robotic policies.

This paper presents an audio-conditioned vision-language-action policy enabling robots to perform manipulation tasks using sound feedback.

This paper proposes estimating kinematic constraints from generative policy ensembles to tune impedance control gains for articulated object manipulation.

This paper uses graph neural networks to predict obstacle avoidance modes, speeding up multimodal model predictive control for autonomous systems.

This paper presents an experimental study protocol measuring how humorous robot apologies affect user trust recovery after robot errors.

This paper introduces a robot manipulation model coordinating semantic reasoning, world modeling, and actions via shared predictive intent tokens.

This paper presents a benchmark evaluating automated repair of real robot navigation software bugs using physics-simulation behavioral oracles.

This paper introduces an industrial benchmark for evaluating long-horizon robotic assembly across diverse DIN-norm compliant manipulation tasks.

This paper proposes an on-policy adaptation framework to adjust legged robot locomotion policies when robot foot morphology or footwear changes.

This paper proposes a dual-branch generative framework that separates surface texture from contact force signals in vision-based tactile sensors.

This paper introduces a geometric framework to model aerodynamic promptness and control authority for tilting-propeller multirotor drones.

This paper presents a nonlinear knee-ankle parallel mechanism coupling that shifts actuator mass closer to the hip in humanoid legs.

The paper evaluates generalization limits in world action models by decoupling future image generation from an inverse-dynamics action recovery model.

The paper presents a framework combining physical validation, finite element simulations, and graph neural networks for real-time soft tactile deformation modeling.

The paper formulates risk-constrained temporal logic planning in Gaussian belief spaces as a sound and complete mixed-integer convex program.

The paper presents a redundancy-resolution method for bimanual humanoid teleoperation combining adaptive elbow guidance with singularity-robust wrist tracking.

The paper introduces a predictive joint-feasibility margin to prevent conflicting acceleration limits in second-order control barrier function quadratic programs.

The paper introduces a gradient-free inference-time control method that adapts diffusion model noise distributions to meet robotic execution constraints.

The paper proposes a task-space motion-force nonlinear model predictive control framework designed to generalize across robotic platforms and unmodeled interactions.

The paper demonstrates that language model coding agents can generate reactive and declarative programs for continuous robot behaviors using targeted guidelines.

The paper presents a navigation framework for humanoids that couples LiDAR safe corridors with footstep planning to negotiate dynamic, non-convex obstacles.

The paper proposes a physics-aware simultaneous localization and mapping method that models dynamic deformable objects using 4D Gaussian fields and spring-mass mechanics.

The paper presents a toolkit combining a neural network and keypoint tracking for dynamic pedestrian re-identification and future occupancy forecasting.

The paper investigates expert intervention delays in robot imitation learning and introduces a method to learn cost functions from delayed interventions.

The paper develops reach-avoid control certificates and safety filters over an infinite horizon for contact-based stochastic hybrid dynamical systems.

The paper presents a benchmark evaluating household robot planners across long horizons, uncertainty, and varied initial memory states.

The paper proposes a control barrier function filter that makes minimal adjustments to guarantee recurrent visits to target sets within specific deadlines.

The paper evaluates a frozen frontier vision-language model used directly as a robot policy on long-horizon memory-dependent manipulation tasks.

The paper introduces a framework to learn coupled observation-action symmetries from offline robot interaction data using adversarial tuple matching.

The paper audits trajectory-smoothness curation metrics and finds they reflect episode length rather than task success in imitation learning datasets.

The paper presents a framework to predict 3D scene flow conditioned on actions using 2D observations for kinematics-free robot control.