archive
Every paper Pith has read. Search by title, abstract, or pith.
2900 papers in cs.RO · page 13
-
Robot checks farm soil moisture using grid sensors and aerial images
Automated Robotic Moisture Monitoring in Agricultural Fields
-
Terminal cost raises success rate in uncertain kinodynamic planning
Terminal Matters: Kinodynamic Planning with a Terminal Cost and Learned Uncertainty in Belief State-Cost Space
-
Terminal cost optimizes path and goal quality in kinodynamic planning
Terminal Matters: Kinodynamic Planning with a Terminal Cost and Learned Uncertainty in Belief State-Cost Space
-
Backdoor lets owners verify adapted VLA models
Towards Backdoor-Based Ownership Verification for Vision-Language-Action Models
-
Adaptive controller tracks laser paths at 2.4 mm in uncalibrated mockup
A low-cost mockup to simulate robotic laser cutting in nuclear decommissioning
-
Raycasting separates dynamic points from static 3D maps
Raymoval: Raycasting-based Dynamic Object Removal for Static 3D Mapping
-
Hyperbolic memory tree lifts VLA long-task success by 12.8 points
ECHO: Continuous Hierarchical Memory for Vision-Language-Action Models
-
Network impairments cut surgical teleoperation success to 12%
VISTA: A Benchmark for Real-Time Video Streaming under Network Impairments in Surgical Teleoperation
-
Confidence scaling keeps VLA skills intact during fine-tuning
Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT
-
ConSFT retains robot skills while adapting to new tasks
Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT
-
Multi-agent framework turns natural language into assembly plans
AssemPlanner: A Multi-Agent Based Task Planning Framework for Flexible Assembly System
-
Expert routing in shared transformer lifts driving score to 88.91
VECTOR-Drive: Tightly Coupled Vision-Language and Trajectory Expert Routing for End-to-End Autonomous Driving
-
Shared attention with expert routing lifts driving performance
VECTOR-Drive: Tightly Coupled Vision-Language and Trajectory Expert Routing for End-to-End Autonomous Driving
-
Diffusion models replace GANs to scale quadruped training
Constraint-Aware Diffusion Priors for High-Fidelity and Versatile Quadruped Locomotion
-
Diffusion prior scales quadruped skills to mixed datasets
Constraint-Aware Diffusion Priors for High-Fidelity and Versatile Quadruped Locomotion
-
One-step policy delivers 71Hz robot control
ElasticFlow: One-Step Physics-Consistent Policy with Elastic Time Horizons for Language-Guided Manipulation
-
ProcVLM reasons about remaining steps before scoring robot progress
ProcVLM: Learning Procedure-Grounded Progress Rewards for Robotic Manipulation
-
Hybrid framework cuts tote movements 8-30% in robot warehouses
Omni-scale Learning-based Sequential Decision Framework for Order Fulfillment of Tote-handling Robotic Systems
-
Direct-drive gripper with button annotation yields structured manipulation demos
A Visuo-Tactile Data Collection System with Haptic Feedback for Coarse-to-Fine Imitation Learning
-
Latent geometry replaces online planning with direct inference
Latent Geometry Beyond Search: Amortizing Planning in World Models
-
Hierarchical coordination scales robot teams to uncertain tasks
HULK: Large-scale Hierarchical Coordination under Continual and Uncertain Temporal Tasks
-
End-to-end RL parks cars in narrow mechanical slots after 3DGS training
REAP: Reinforcement-Learning End-to-End Autonomous Parking with Gaussian Splatting Simulator for Real2Sim2Real Transfer
-
Clustering action chunks raises robot task success by 13.3%
Geometry Guided Self-Consistency for Physical AI
-
Adaptive tuning enables 80% success backdoor attacks on VLA models at 5% poisoning
ATAAT: Adaptive Threat-Aware Adversarial Tuning Framework against Backdoor Attacks on Vision-Language-Action Models
-
Reweighting source examples boosts robot policy performance across domains
BEACON: Cross-Domain Co-Training of Generative Robot Policies via Best-Effort Adaptation
-
BEACON reweights source robot data to improve target-domain policy performance
BEACON: Cross-Domain Co-Training of Generative Robot Policies via Best-Effort Adaptation
-
GPU vectorization speeds physics driving sims 127-fold
SceneFactory: GPU-Accelerated Multi-Agent Driving Simulation with Physics-Based Vehicle Dynamics
-
Adaptive control enables precise flight for 95-mg bee robot
Model-Reference Adaptive Flight Control of the 95-mg Bee++
-
Quantile sorting makes flow matching Wasserstein-aligned for DRL
Quantile-Coupled Flow Matching for Distributional Reinforcement Learning
-
Flow matching policies reach 70% on tasks where baselines score 0%
Trajectory-Consistent Flow Matching for Robust Visuomotor Policy Learning
-
Hybrid model cuts vehicle prediction errors up to 91 percent
LE-PAVD: Learning-Enhanced Physics-Aware Vehicle Dynamics for High-Speed Autonomous Navigation
-
Failure rollouts steer VLA policies toward higher success
Failing Forward: Adaptive Failure-Informed Learning for Vision-Language-Action Models
-
Online failures train better VLA policies for robots
Failing Forward: Adaptive Failure-Informed Learning for Vision-Language-Action Models
-
One API unifies 3300 hours of driving data across eight datasets
123D: Unifying Multi-Modal Autonomous Driving Data at Scale
-
RGB-D cross-fusion lifts 6D pose accuracy to 92% on LINEMOD
6D Pose Estimation via Keypoint Heatmap Regression with RGB-D Residual Neural Networks
-
Dual-LLM hierarchy drives 86% robot planning success
Hierarchical Prompting with Dual LLM Modules for Robotic Task and Motion Planning
-
Legged robots learn to identify their own body parameters through active RL
Active Embodiment Identification with Reinforcement Learning for Legged Robots
-
Actuated spine boosts quadruped agility in simulation
Evaluation of an Actuated Spine in Agile Quadruped Locomotion
-
Active vision improves imitation learning on some tasks
TAVIS: A Benchmark for Egocentric Active Vision and Anticipatory Gaze in Imitation Learning
-
Event-based SLAM delivers first onboard closed-loop UAV control
AERO-VIS: Asynchronous Event-based Real-time Onboard Visual-Inertial SLAM
-
Logic rules steer LLMs to create valid robot swarm plans
Melding LLM and temporal logic for reliable human-swarm collaboration in complex scenarios
-
M2M method finishes up to 22,000 more warehouse tasks than prior approaches
Many-to-Many Multi-Agent Pickup and Delivery
-
Executable tests benchmark Text-to-CAD and beat existing methods
Text-to-CAD Evaluation with CADTests
-
Learned per-frame noise levels improve robot actions
NoiseGate: Learning Per-Latent Timestep Schedules as Information Gating in World Action Models
-
-
LoRA-adapted LLM turns text into valid swarm behaviors
CommandSwarm: Safety-Aware Natural Language-to-Behavior-Tree Generation for Robotic Swarms
-
Synthetic LiDAR descriptors relocalize robots in 3 seconds
Offline-Online Hierarchical 3D Global Relocalization With Synthetic LiDAR Sensing and Descriptor-Space Retrieval
-
One-step policy via Wasserstein flow beats ODE baselines on robot tasks
Drifting Field Policy: A One-Step Generative Policy via Wasserstein Gradient Flow
-
Voro-POMCPOW gives finite-time bounds for continuous POMDP planning
Finite-Time Analysis of MCTS in Continuous POMDP Planning
-
GNN reduces spring-mass twins up to 2.3 times faster
PhySPRING: Structure-Preserving Reduction of Physics-Informed Twins via GNN