Ping Luo

A 4B grounding foundation model achieves state-of-the-art visual grounding, outperforming larger models and boosting downstream robotic and autonomous driving performance by up to 24.8% in OOD settings.

This work introduces DexTouch-WM, an action-conditioned world model that learns from scalable human touch to jointly predict future RGB observations and bilateral tactile dynamics, and shows that scalable human interaction provides a complementary data axis for learning dexterous robot world models.

This work takes an RSI-inspired approach at the harness layer, scaling auto-research loops across increasingly numerous and diverse environments for harness rollouts, and yields reusable improvements that transfer beyond their development setting.

Treating distillation targets as dynamic on-policy decisions rather than static teacher outputs prevents catastrophic error propagation when adapting compact vision-language models to out-of-distribution multimodal data.

Policies may succeed in tasks but still violate deformation tolerances, revealing a critical gap in current evaluation methods for deformable-object manipulation.

Integration of diverse robot policies can be streamlined from hours to minutes, revolutionizing how we deploy and evaluate robotic systems.

Trustworthiness in embodied intelligence isn't just about performance; it's about managing risk across a multi-layered framework that ensures safety and reliability in real-world applications.

The data pyramid framework reveals how the interplay of diverse data sources can unlock new capabilities in embodied agents, highlighting critical gaps in current methodologies.

Many robotic policies that seem successful in manipulation tasks actually compromise safety, with SoftVTBench revealing a stark contrast between goal completion and physical safety metrics.

RoboDojo reveals that integrating simulation and real-world tasks can significantly enhance the evaluation of robot manipulation policies, bridging the gap between theoretical performance and practical deployment.

SR-REAL's dual-path reasoning framework allows spatial VLMs to excel in both linguistic deduction and 3D geometric inference, significantly enhancing performance on complex spatial reasoning tasks.

MemoryVLA++ achieves up to 28% performance gains in robotic manipulation tasks by integrating memory and imagination, transforming how robots handle temporal dependencies.

Achieve near-perfect sim-to-real transfer in robotic manipulation by closing the loop between synthetic data generation and robust policy training.

Decoupling high-level VLM planning from low-level diffusion-based control lets robots reason like foundation models *and* execute precisely, outperforming end-to-end approaches in complex manipulation tasks.

Forget painstakingly creating 3D assets for robot training - ManiTwin automates the process, turning single images into simulation-ready objects at scale.

Forget slow per-scene optimization: ReconDrive uses a fast feed-forward approach to generate high-fidelity 4D Gaussian Splatting for autonomous driving, rivaling optimization-based methods in quality while being orders of magnitude faster.

Forget short-term context windows: VPWEM's Transformer-based memory compressor lets robots ace long-horizon manipulation tasks by distilling past observations into fixed-size episodic memories.