Yunzhu Li

Cheng Qian, Emre Can Acikgoz, Bingxuan Li et al.

Agents built on vision-language models increasingly face tasks that demand anticipating future states rather than relying on short-horizon reasoning. Generative world models offer a promising remedy: agents could use them as external simulators to foresee outcomes before acting. This paper empirically examines whether current agents can leverage such world models as tools to enhance their cognition. Across diverse agentic and visual question answering tasks, we observe that some agents rarely invoke simulation (fewer than 1%), frequently misuse predicted rollouts (approximately 15%), and often exhibit inconsistent or even degraded performance (up to 5%) when simulation is available or enforced. Attribution analysis further indicates that the primary bottleneck lies in the agents' capacity to decide when to simulate, how to interpret predicted outcomes, and how to integrate foresight into downstream reasoning. These findings underscore the need for mechanisms that foster calibrated, strategic interaction with world models, paving the way toward more reliable anticipatory cognition in future agent systems.

Ziang Liu, Genggeng Zhou, Jeff He et al.

Learning predictive models from observations using deep neural networks (DNNs) is a promising new approach to many real-world planning and control problems. However, common DNNs are too unstructured for effective planning, and current control methods typically rely on extensive sampling or local gradient descent. In this paper, we propose a new framework for integrated model learning and predictive control that is amenable to efficient optimization algorithms. Specifically, we start with a ReLU neural model of the system dynamics and, with minimal losses in prediction accuracy, we gradually sparsify it by removing redundant neurons. This discrete sparsification process is approximated as a continuous problem, enabling an end-to-end optimization of both the model architecture and the weight parameters. The sparsified model is subsequently used by a mixed-integer predictive controller, which represents the neuron activations as binary variables and employs efficient branch-and-bound algorithms. Our framework is applicable to a wide variety of DNNs, from simple multilayer perceptrons to complex graph neural dynamics. It can efficiently handle tasks involving complicated contact dynamics, such as object pushing, compositional object sorting, and manipulation of deformable objects. Numerical and hardware experiments show that, despite the aggressive sparsification, our framework can deliver better closed-loop performance than existing state-of-the-art methods.

Yunzhu Li, Hao He, Jiajun Wu et al.

Finding an embedding space for a linear approximation of a nonlinear dynamical system enables efficient system identification and control synthesis. The Koopman operator theory lays the foundation for identifying the nonlinear-to-linear coordinate transformations with data-driven methods. Recently, researchers have proposed to use deep neural networks as a more expressive class of basis functions for calculating the Koopman operators. These approaches, however, assume a fixed dimensional state space; they are therefore not applicable to scenarios with a variable number of objects. In this paper, we propose to learn compositional Koopman operators, using graph neural networks to encode the state into object-centric embeddings and using a block-wise linear transition matrix to regularize the shared structure across objects. The learned dynamics can quickly adapt to new environments of unknown physical parameters and produce control signals to achieve a specified goal. Our experiments on manipulating ropes and controlling soft robots show that the proposed method has better efficiency and generalization ability than existing baselines.

Yunzhu Li, Jiajun Wu, Jun-Yan Zhu et al.

There has been an increasing interest in learning dynamics simulators for model-based control. Compared with off-the-shelf physics engines, a learnable simulator can quickly adapt to unseen objects, scenes, and tasks. However, existing models like interaction networks only work for fully observable systems; they also only consider pairwise interactions within a single time step, both restricting their use in practical systems. We introduce Propagation Networks (PropNet), a differentiable, learnable dynamics model that handles partially observable scenarios and enables instantaneous propagation of signals beyond pairwise interactions. Experiments show that our propagation networks not only outperform current learnable physics engines in forward simulation, but also achieve superior performance on various control tasks. Compared with existing model-free deep reinforcement learning algorithms, model-based control with propagation networks is more accurate, efficient, and generalizable to new, partially observable scenes and tasks.

Yunzhu Li, Jiaming Song, Stefano Ermon

The goal of imitation learning is to mimic expert behavior without access to an explicit reward signal. Expert demonstrations provided by humans, however, often show significant variability due to latent factors that are typically not explicitly modeled. In this paper, we propose a new algorithm that can infer the latent structure of expert demonstrations in an unsupervised way. Our method, built on top of Generative Adversarial Imitation Learning, can not only imitate complex behaviors, but also learn interpretable and meaningful representations of complex behavioral data, including visual demonstrations. In the driving domain, we show that a model learned from human demonstrations is able to both accurately reproduce a variety of behaviors and accurately anticipate human actions using raw visual inputs. Compared with various baselines, our method can better capture the latent structure underlying expert demonstrations, often recovering semantically meaningful factors of variation in the data.

Yunzhu Li, Benyuan Sun, Tianfu Wu et al.

This paper presents a method for face detection in the wild, which integrates a ConvNet and a 3D mean face model in an end-to-end multi-task discriminative learning framework. The 3D mean face model is predefined and fixed (e.g., we used the one provided in the AFLW dataset). The ConvNet consists of two components: (i) The face pro- posal component computes face bounding box proposals via estimating facial key-points and the 3D transformation (rotation and translation) parameters for each predicted key-point w.r.t. the 3D mean face model. (ii) The face verification component computes detection results by prun- ing and refining proposals based on facial key-points based configuration pooling. The proposed method addresses two issues in adapting state- of-the-art generic object detection ConvNets (e.g., faster R-CNN) for face detection: (i) One is to eliminate the heuristic design of prede- fined anchor boxes in the region proposals network (RPN) by exploit- ing a 3D mean face model. (ii) The other is to replace the generic RoI (Region-of-Interest) pooling layer with a configuration pooling layer to respect underlying object structures. The multi-task loss consists of three terms: the classification Softmax loss and the location smooth l1 -losses [14] of both the facial key-points and the face bounding boxes. In ex- periments, our ConvNet is trained on the AFLW dataset only and tested on the FDDB benchmark with fine-tuning and on the AFW benchmark without fine-tuning. The proposed method obtains very competitive state-of-the-art performance in the two benchmarks.