Xuelong Li

By pretraining a VLA model with goal-conditioned RL, PRTS learns to reason about goal reachability, leading to substantial gains in long-horizon robotic tasks and zero-shot generalization.

Humanoid robots can now seamlessly transition between fighting skills thanks to a novel policy gating approach that ensures stability and smoothness.

Humanoid robots can now adapt to diverse environments without task-specific tuning by selectively "relaxing" joints, mimicking how humans exploit weightlessness for stability.

Object hallucination in LVLMs can be significantly reduced *after* training, without any extra data or compute.

Stop blasting your diffusion models with a single, static reward signal: fine-grained credit assignment across denoising steps and objectives unlocks better image and video generation.

Ditch the learned router: a global scheduler for Mixture-of-Experts models unlocks state-of-the-art multi-domain learning by explicitly optimizing dataset-to-expert assignments.

Finetuning visual foundation models with LoRA-based pairwise training dramatically improves AIGI detection robustness against real-world distortions.

Forget static coordination – robots that chat and dynamically re-plan can achieve a whopping 69% improvement in collaborative navigation success.

Single-pixel imaging gets a deep learning boost: SISTA-Net leverages learned sparsity and hybrid CNN-VSSM architectures to achieve state-of-the-art reconstruction quality, even in noisy underwater environments.

Ruyi2.5 achieves comparable performance to Qwen3-VL on general multimodal benchmarks while significantly outperforming it in privacy-constrained surveillance, demonstrating the effectiveness of its edge-cloud architecture.

Achieve real-time (409 FPS) underwater image enhancement with a tiny (3,880 parameter) model that significantly improves color accuracy, enabling deployment on resource-constrained underwater platforms.

Don't fully retrain your draft model after fine-tuning your LLM: EDA restores speculative decoding performance with significantly less compute by adapting only a small, private component and regenerating training data.

Skip the costly robot teleoperation data: ZeroWBC learns surprisingly natural humanoid control policies directly from human egocentric videos.

LLMs can learn better from human feedback by exploring more creatively, thanks to a simple coin-flip counting method that encourages them to try new things.