icra2026

Official PyTorch implementation of Superpoint Transformer [ICCV'23], SuperCluster [3DV'24 Oral], and EZ-SP [ICRA'26]

Official code for "Reinforcement Fine-Tuning of Flow-Matching Policies for Vision-Language-Action Models" (ICRA 2026): Flow Policy Optimization for pi0

[ICRA 2026] Official implementation of OTAS: Open-vocabulary Token Alignment for Outdoor Segmentation. Training-free language-grounded environment reconstruction and semantic segmentation. Built for in the wild, real-time robotic applications in unstructured and uncooperative environments.

[ICRA 2026] Official implementation of the paper: "TagaVLM: Topology-Aware Global Action Reasoning for Vision-Language Navigation"

Dexora: Open-source VLA for High-DoF Bimanual Dexterity (ICRA 2026). Dual-arm, dual-hand, 36-DoF diffusion-transformer policy with discriminator-guided data-quality-aware post-training.

PolygMap: A Perceptive Locomotion Framework for Humanoid Robot Stair Climbing

CMAR-search: Commonsense and Memory Augmented Reasoning for Object Search in Dynamic Interactive Environments (ICRA 2026)

ICRA 2026 | T2M Mamba: Motion Periodicity-Saliency Coupling Approach for Stable Text-Driven Motion Generation.

Per-Group Error, Not Total MSE. Fine-tuning VLAs on the Toyota HSR. ICRA 2026 Workshop From Data to Decisions, 1st place out of 36.

BOP-style benchmark for object grasping — ICRA 2026

MANIP-RAAP — Affordance Prediction

Imitation-learning policies for the Leju Kuavo humanoid in the Kuavo Data Challenge @ ICRA 2026 (simulation track). ACT trained on 1,000 demos per task across toy sorting, parcel weighing and conveyor sorting: 258.0/300 on local BENCH, top-10 level. Includes training, MuJoCo evaluation harness, results and full reproduction.

My entry for the AgiBot World Challenge @ ICRA 2026, world model track. GE-Sim 2.0 fine-tuned with a rank-64 LoRA turns one real camera frame plus the robot’s actions into the video of what happens next. Validation score 0.8079.

ERL poster contribution to IEEE ICRA 2026.

A full autonomous racing stack built on ROS 2 Humble for the AutoDRIVE RoboRacer platform.

ScaleMaster Office 01 (ICRA 2026, DGIST APRL): Gaussian splatting, Dalgu walk mode and guided tour.