MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation

Hao Shi, Bin Xie, Yingfei Liu, Lin Sun, Fengrong Liu, Tiancai Wang, Erjin Zhou, Haoqiang Fan, Xiangyu Zhang, Gao Huang✉

• MemoryVLA is a Cognition-Memory-Action framework for robotic manipulation inspired by human memory systems. It builds a hippocampal-like perceptual-cognitive memory to capture the temporal dependencies essential for current decision-making, enabling long-horizon, temporally aware action generation.

DenseGrounding: Improving Dense Language-Vision Semantics for Ego-centric 3D Visual Grounding

Henry Zheng*, Hao Shi*, Qihang Peng, Yong Xien Chng, Rui Huang, Yepeng Weng, Zhongchao Shi, Gao Huang#✉

• DenseGrounding is a framework for embodied 3D visual grounding. It tackles the loss of fine-grained visual semantics from sparse fusion of point clouds with multi-view images, as well as the limited textual semantic context from arbitrary instructions, enabling more accurate and context-aware grounding.

SpatialActor: Exploring Disentangled Spatial Representations for Robust Robotic Manipulation

Hao Shi, Bin Xie, Yingfei Liu, Yang Yue, Tiancai Wang, Haoqiang Fan, Xiangyu Zhang, Gao Huang✉

• SpatialActor is a disentangled framework for robust robotic manipulation. It decouples perception into complementary high-level geometry from fine-grained but noisy raw depth and coarse but robust depth expert priors, along with low-level spatial cues and appearance semantics.

DenseG: Alleviating Vision-Language Feature Sparsity in Multi-View 3D Visual Grounding

Henry Zheng*, Hao Shi*, Yong Xien Chng, Rui Huang, Zanlin Ni, Tianyi Tan, Qihang Peng, Yepeng Weng, Zhongchao Shi, Gao Huang✉

• 1st Place and Innovation Award in CVPR 2024 Autonomous Grand Challenge, Embodied 3D Grounding Track (1/64 teams, 1/154 submissions).
• Oral Presentation at CVPR 2024 Workshop on Foundation Models for Autonomous Systems.

Open Compound Domain Adaptation with Object Style Compensation for Semantic Segmentation

Tingliang Feng*, Hao Shi*, Xueyang Liu, Wei Feng, Liang Wan, Yanlin Zhou, Di Lin✉

• We propose object style compensation for open compound domain adaptation. It builds an object-level discrepancy memory bank to capture fine-grained source–target domain gaps and compensates target features to align with source distribution, enabling cross-domain segmentation.

GeoVLA: Enpowering 3D Representations in Vision-Language-Action Models

Lin Sun*, Bin Xie*, Yingfei Liu, Hao Shi, Tiancai Wang, Jiale Cao✉

• GeoVLA is a framework that bridges 2D semantics and 3D geometry for VLA. By encoding geometric embeddings with a dual-stream design and leveraging a Mixture-of-Experts 3D-Aware Action Expert, it achieves robustness across diverse camera views, object heights, and sizes.

Grounding Beyond Detection: Enhancing Contextual Understanding in Embodied 3D Grounding

Yani Zhang*, Dongming Wu*, Hao Shi, Yingfei Liu, Tiancai Wang, Haoqiang Fan, Xingping Dong✉

• DEGround transfers general proposals from detection into grounding with shared queries, and mitigates vision–language misalignment through region activation and query-wise modulation, achieving 1st place on EmbodiedScan.