Robotics

Zero-shot Neural ODEs enable autonomous systems to rapidly adapt to unseen scenarios by learning a space of dynamical systems spanned by neural ODE basis functions, achieving efficient online adaptati…

VLMimic: Vision-Language Models enable robots to master intricate actions using only a few human video demonstrations, surpassing existing methods by a significant margin.

VidMan: a novel framework leverages video diffusion models and a two-stage training mechanism to significantly improve robot manipulation precision by effectively using robot trajectory data and impli…

VDD distills complex diffusion policies into efficient Mixture of Experts (MoE) models via variational inference, enabling faster inference and improved performance in behavior learning.

TrajCLIP: a novel pedestrian trajectory prediction method using contrastive learning and idempotent networks to achieve state-of-the-art performance and enhance generalization across diverse scenarios…

Team-Fictitious Play (Team-FP) enables self-interested agents to learn near-optimal team coordination in multi-team games, reaching a Team-Nash equilibrium with quantifiable error bounds.

Human understanding of formal specifications for robot validation is surprisingly poor; active learning, while improving engagement, doesn’t significantly boost accuracy.

IDA, a novel intervention assistance, dynamically shares control between human and AI copilots by intervening only when the AI’s action is superior across all goals, maximizing performance and preserv…

SCaR refines skill chaining for long-horizon robotic manipulation via dual regularization, achieving higher success rates and robustness.

Heterogeneous Pre-trained Transformers (HPT) enables robots to learn generalizable policies from diverse data, drastically improving performance on unseen tasks.

TVSAFEOPT: Safe time-varying optimization using spatio-temporal kernels ensures safety while tracking time-varying reward and safety functions, providing optimality guarantees in stationary settings.

RL-GPT seamlessly integrates Large Language Models (LLMs) and Reinforcement Learning (RL) to create highly efficient agents mastering complex tasks in open-world environments.

QueST: A novel self-supervised skill abstraction architecture for continuous robot control, achieves state-of-the-art performance on multitask and few-shot learning benchmarks by learning flexible, tr…

PAD, a novel visual policy learning framework, unifies image prediction and robot action in a joint denoising process, achieving significant performance improvements in robotic manipulation tasks.

Predicting RL agent behavior is key for safety and interaction; this study reveals that explicitly planned agents are significantly easier to predict due to their internal plans.

Omnigrasp: A novel RL-based method enables simulated humanoids to grasp diverse objects and precisely follow complex trajectories, advancing realistic human-object interaction in virtual environments.

NeuralSteiner uses deep learning to predict Steiner points for efficient, overflow-avoiding global routing in chip design, achieving up to a 99.8% overflow reduction on large benchmarks.

NeuralFluid: Design & control complex fluidic systems with dynamic boundaries using differentiable simulation, achieving superior results in benchmark tasks.

Model-Based Diffusion (MBD) uses diffusion processes and model information for data-free trajectory optimization, outperforming existing methods on complex tasks.

MGF: Mixed Gaussian Flow enhances trajectory prediction by using a mixed Gaussian prior, achieving state-of-the-art diversity and alignment accuracy.