Yuqing Du

Yuqing Du, Olivia Watkins, Zihan Wang et al. · microsoft-research, mit

Reinforcement learning algorithms typically struggle in the absence of a dense, well-shaped reward function. Intrinsically motivated exploration methods address this limitation by rewarding agents for visiting novel states or transitions, but these methods offer limited benefits in large environments where most discovered novelty is irrelevant for downstream tasks. We describe a method that uses background knowledge from text corpora to shape exploration. This method, called ELLM (Exploring with LLMs) rewards an agent for achieving goals suggested by a language model prompted with a description of the agent's current state. By leveraging large-scale language model pretraining, ELLM guides agents toward human-meaningful and plausibly useful behaviors without requiring a human in the loop. We evaluate ELLM in the Crafter game environment and the Housekeep robotic simulator, showing that ELLM-trained agents have better coverage of common-sense behaviors during pretraining and usually match or improve performance on a range of downstream tasks. Code available at https://github.com/yuqingd/ellm.

Jessy Lin, Yuqing Du, Olivia Watkins et al. · deepmind, utoronto

To interact with humans and act in the world, agents need to understand the range of language that people use and relate it to the visual world. While current agents can learn to execute simple language instructions, we aim to build agents that leverage diverse language -- language like "this button turns on the TV" or "I put the bowls away" -- that conveys general knowledge, describes the state of the world, provides interactive feedback, and more. Our key idea is that agents should interpret such diverse language as a signal that helps them predict the future: what they will observe, how the world will behave, and which situations will be rewarded. This perspective unifies language understanding with future prediction as a powerful self-supervised learning objective. We instantiate this in Dynalang, an agent that learns a multimodal world model to predict future text and image representations, and learns to act from imagined model rollouts. While current methods that learn language-conditioned policies degrade in performance with more diverse types of language, we show that Dynalang learns to leverage environment descriptions, game rules, and instructions to excel on tasks ranging from game-playing to navigating photorealistic home scans. Finally, we show that our method enables additional capabilities due to learning a generative model: Dynalang can be pretrained on text-only data, enabling learning from offline datasets, and generate language grounded in an environment.

Kimin Lee, Hao Liu, Moonkyung Ryu et al.

Deep generative models have shown impressive results in text-to-image synthesis. However, current text-to-image models often generate images that are inadequately aligned with text prompts. We propose a fine-tuning method for aligning such models using human feedback, comprising three stages. First, we collect human feedback assessing model output alignment from a set of diverse text prompts. We then use the human-labeled image-text dataset to train a reward function that predicts human feedback. Lastly, the text-to-image model is fine-tuned by maximizing reward-weighted likelihood to improve image-text alignment. Our method generates objects with specified colors, counts and backgrounds more accurately than the pre-trained model. We also analyze several design choices and find that careful investigations on such design choices are important in balancing the alignment-fidelity tradeoffs. Our results demonstrate the potential for learning from human feedback to significantly improve text-to-image models.

Yuqing Du, Ksenia Konyushkova, Misha Denil et al.

Detecting successful behaviour is crucial for training intelligent agents. As such, generalisable reward models are a prerequisite for agents that can learn to generalise their behaviour. In this work we focus on developing robust success detectors that leverage large, pretrained vision-language models (Flamingo, Alayrac et al. (2022)) and human reward annotations. Concretely, we treat success detection as a visual question answering (VQA) problem, denoted SuccessVQA. We study success detection across three vastly different domains: (i) interactive language-conditioned agents in a simulated household, (ii) real world robotic manipulation, and (iii) "in-the-wild" human egocentric videos. We investigate the generalisation properties of a Flamingo-based success detection model across unseen language and visual changes in the first two domains, and find that the proposed method is able to outperform bespoke reward models in out-of-distribution test scenarios with either variation. In the last domain of "in-the-wild" human videos, we show that success detection on unseen real videos presents an even more challenging generalisation task warranting future work. We hope our initial results encourage further work in real world success detection and reward modelling.

Philipp Wu, Kourosh Hakhamaneshi, Yuqing Du et al.

One-shot Imitation Learning~(OSIL) aims to imbue AI agents with the ability to learn a new task from a single demonstration. To supervise the learning, OSIL typically requires a prohibitively large number of paired expert demonstrations -- i.e. trajectories corresponding to different variations of the same semantic task. To overcome this limitation, we introduce the semi-supervised OSIL problem setting, where the learning agent is presented with a large dataset of trajectories with no task labels (i.e. an unpaired dataset), along with a small dataset of multiple demonstrations per semantic task (i.e. a paired dataset). This presents a more realistic and practical embodiment of few-shot learning and requires the agent to effectively leverage weak supervision from a large dataset of trajectories. Subsequently, we develop an algorithm specifically applicable to this semi-supervised OSIL setting. Our approach first learns an embedding space where different tasks cluster uniquely. We utilize this embedding space and the clustering it supports to self-generate pairings between trajectories in the large unpaired dataset. Through empirical results on simulated control tasks, we demonstrate that OSIL models trained on such self-generated pairings are competitive with OSIL models trained with ground-truth labels, presenting a major advancement in the label-efficiency of OSIL.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Aly Lidayan, Yuqing Du, Eliza Kosoy et al.

What drives exploration? Understanding intrinsic motivation is a long-standing challenge in both cognitive science and artificial intelligence; numerous objectives have been proposed and used to train agents, yet there remains a gap between human and agent exploration. We directly compare adults, children, and AI agents in a complex open-ended environment, Crafter, and study how common intrinsic objectives: Entropy, Information Gain, and Empowerment, relate to their behavior. We find that only Entropy and Empowerment are consistently positively correlated with human exploration progress, indicating that these objectives may better inform intrinsic reward design for agents. Furthermore, across agents and humans we observe that Entropy initially increases rapidly, then plateaus, while Empowerment increases continuously, suggesting that state diversity may provide more signal in early exploration, while advanced exploration should prioritize control. Finally, we find preliminary evidence that private speech utterances, and particularly goal verbalizations, may aid exploration in children. Our data is available at https://github.com/alyd/humans_in_crafter_data.

Imagen-Team-Google, Jason Baldridge, Jakob Bauer et al.

We introduce Imagen 3, a latent diffusion model that generates high quality images from text prompts. We describe our quality and responsibility evaluations. Imagen 3 is preferred over other state-of-the-art (SOTA) models at the time of evaluation. In addition, we discuss issues around safety and representation, as well as methods we used to minimize the potential harm of our models.

Alex Havrilla, Yuqing Du, Sharath Chandra Raparthy et al. · meta-ai

Reinforcement Learning from Human Feedback (\textbf{RLHF}) has emerged as a dominant approach for aligning LLM outputs with human preferences. Inspired by the success of RLHF, we study the performance of multiple algorithms that learn from feedback (Expert Iteration, Proximal Policy Optimization (\textbf{PPO}), Return-Conditioned RL) on improving LLM reasoning capabilities. We investigate both sparse and dense rewards provided to the LLM both heuristically and via a learned reward model. We additionally start from multiple model sizes and initializations both with and without supervised fine-tuning (\textbf{SFT}) data. Overall, we find all algorithms perform comparably, with Expert Iteration performing best in most cases. Surprisingly, we find the sample complexity of Expert Iteration is similar to that of PPO, requiring at most on the order of $10^6$ samples to converge from a pretrained checkpoint. We investigate why this is the case, concluding that during RL training models fail to explore significantly beyond solutions already produced by SFT models. Additionally, we discuss a trade off between maj@1 and pass@96 metric performance during SFT training and how conversely RL training improves both simultaneously. We then conclude by discussing the implications of our findings for RLHF and the future role of RL in LLM fine-tuning.

Ying Fan, Olivia Watkins, Yuqing Du et al.

Learning from human feedback has been shown to improve text-to-image models. These techniques first learn a reward function that captures what humans care about in the task and then improve the models based on the learned reward function. Even though relatively simple approaches (e.g., rejection sampling based on reward scores) have been investigated, fine-tuning text-to-image models with the reward function remains challenging. In this work, we propose using online reinforcement learning (RL) to fine-tune text-to-image models. We focus on diffusion models, defining the fine-tuning task as an RL problem, and updating the pre-trained text-to-image diffusion models using policy gradient to maximize the feedback-trained reward. Our approach, coined DPOK, integrates policy optimization with KL regularization. We conduct an analysis of KL regularization for both RL fine-tuning and supervised fine-tuning. In our experiments, we show that DPOK is generally superior to supervised fine-tuning with respect to both image-text alignment and image quality. Our code is available at https://github.com/google-research/google-research/tree/master/dpok.

Yuqing Du, Pieter Abbeel, Aditya Grover

We are interested in training general-purpose reinforcement learning agents that can solve a wide variety of goals. Training such agents efficiently requires automatic generation of a goal curriculum. This is challenging as it requires (a) exploring goals of increasing difficulty, while ensuring that the agent (b) is exposed to a diverse set of goals in a sample efficient manner and (c) does not catastrophically forget previously solved goals. We propose Curriculum Self Play (CuSP), an automated goal generation framework that seeks to satisfy these desiderata by virtue of a multi-player game with four agents. We extend the asymmetric curricula learning in PAIRED (Dennis et al., 2020) to a symmetrized game that carefully balances cooperation and competition between two off-policy student learners and two regret-maximizing teachers. CuSP additionally introduces entropic goal coverage and accounts for the non-stationary nature of the students, allowing us to automatically induce a curriculum that balances progressive exploration with anti-catastrophic exploitation. We demonstrate that our method succeeds at generating an effective curricula of goals for a range of control tasks, outperforming other methods at zero-shot test-time generalization to novel out-of-distribution goals.

Yuqing Du, Daniel Ho, Alexander A. Alemi et al.

In this work we investigate and demonstrate benefits of a Bayesian approach to imitation learning from multiple sensor inputs, as applied to the task of opening office doors with a mobile manipulator. Augmenting policies with additional sensor inputs, such as RGB + depth cameras, is a straightforward approach to improving robot perception capabilities, especially for tasks that may favor different sensors in different situations. As we scale multi-sensor robotic learning to unstructured real-world settings (e.g. offices, homes) and more complex robot behaviors, we also increase reliance on simulators for cost, efficiency, and safety. Consequently, the sim-to-real gap across multiple sensor modalities also increases, making simulated validation more difficult. We show that using the Variational Information Bottleneck (Alemi et al., 2016) to regularize convolutional neural networks improves generalization to held-out domains and reduces the sim-to-real gap in a sensor-agnostic manner. As a side effect, the learned embeddings also provide useful estimates of model uncertainty for each sensor. We demonstrate that our method is able to help close the sim-to-real gap and successfully fuse RGB and depth modalities based on understanding of the situational uncertainty of each sensor. In a real-world office environment, we achieve 96% task success, improving upon the baseline by +16%.

Mohi Khansari, Daniel Ho, Yuqing Du et al.

Recent work in visual end-to-end learning for robotics has shown the promise of imitation learning across a variety of tasks. Such approaches are expensive both because they require large amounts of real world training demonstrations and because identifying the best model to deploy in the real world requires time-consuming real-world evaluations. These challenges can be mitigated by simulation: by supplementing real world data with simulated demonstrations and using simulated evaluations to identify high performing policies. However, this introduces the well-known "reality gap" problem, where simulator inaccuracies decorrelate performance in simulation from that of reality. In this paper, we build on top of prior work in GAN-based domain adaptation and introduce the notion of a Task Consistency Loss (TCL), a self-supervised loss that encourages sim and real alignment both at the feature and action-prediction levels. We demonstrate the effectiveness of our approach by teaching a mobile manipulator to autonomously approach a door, turn the handle to open the door, and enter the room. The policy performs control from RGB and depth images and generalizes to doors not encountered in training data. We achieve 72% success across sixteen seen and unseen scenes using only ~16.2 hours of teleoperated demonstrations in sim and real. To the best of our knowledge, this is the first work to tackle latched door opening from a purely end-to-end learning approach, where the task of navigation and manipulation are jointly modeled by a single neural network.

Yuqing Du, Olivia Watkins, Trevor Darrell et al.

Policies trained in simulation often fail when transferred to the real world due to the `reality gap' where the simulator is unable to accurately capture the dynamics and visual properties of the real world. Current approaches to tackle this problem, such as domain randomization, require prior knowledge and engineering to determine how much to randomize system parameters in order to learn a policy that is robust to sim-to-real transfer while also not being too conservative. We propose a method for automatically tuning simulator system parameters to match the real world using only raw RGB images of the real world without the need to define rewards or estimate state. Our key insight is to reframe the auto-tuning of parameters as a search problem where we iteratively shift the simulation system parameters to approach the real-world system parameters. We propose a Search Param Model (SPM) that, given a sequence of observations and actions and a set of system parameters, predicts whether the given parameters are higher or lower than the true parameters used to generate the observations. We evaluate our method on multiple robotic control tasks in both sim-to-sim and sim-to-real transfer, demonstrating significant improvement over naive domain randomization. Project videos and code at https://yuqingd.github.io/autotuned-sim2real/

Yuqing Du, Nicholas J. Hetherington, Chu Lip Oon et al.

In this paper, we propose a novel navigation system for mobile robots in pedestrian-rich sidewalk environments. Sidewalks are unique in that the pedestrian-shared space has characteristics of both roads and indoor spaces. Like vehicles on roads, pedestrian movement often manifests as linear flows in opposing directions. On the other hand, pedestrians also form crowds and can exhibit much more random movements than vehicles. Classical algorithms are insufficient for safe navigation around pedestrians and remaining on the sidewalk space. Thus, our approach takes advantage of natural human motion to allow a robot to adapt to sidewalk navigation in a safe and socially-compliant manner. We developed a \textit{group surfing} method which aims to imitate the optimal pedestrian group for bringing the robot closer to its goal. For pedestrian-sparse environments, we propose a sidewalk edge detection and following method. Underlying these two navigation methods, the collision avoidance scheme is human-aware. The integrated navigation stack is evaluated and demonstrated in simulation. A hardware demonstration is also presented.

Qunsong Zeng, Yuqing Du, Kaibin Huang

Federated edge learning (FEEL) is a widely adopted framework for training an artificial intelligence (AI) model distributively at edge devices to leverage their data while preserving their data privacy. The execution of a power-hungry learning task at energy-constrained devices is a key challenge confronting the implementation of FEEL. To tackle the challenge, we propose the solution of powering devices using wireless power transfer (WPT). To derive guidelines on deploying the resultant wirelessly powered FEEL (WP-FEEL) system, this work aims at the derivation of the tradeoff between the model convergence and the settings of power sources in two scenarios: 1) the transmission power and density of power-beacons (dedicated charging stations) if they are deployed, or otherwise 2) the transmission power of a server (access-point). The development of the proposed analytical framework relates the accuracy of distributed stochastic gradient estimation to the WPT settings, the randomness in both communication and WPT links, and devices' computation capacities. Furthermore, the local-computation at devices (i.e., mini-batch size and processor clock frequency) is optimized to efficiently use the harvested energy for gradient estimation. The resultant learning-WPT tradeoffs reveal the simple scaling laws of the model-convergence rate with respect to the transferred energy as well as the devices' computational energy efficiencies. The results provide useful guidelines on WPT provisioning to provide a guaranteer on learning performance. They are corroborated by experimental results using a real dataset.

Eugene Vinitsky, Yuqing Du, Kanaad Parvate et al.

Reinforcement Learning (RL) is an effective tool for controller design but can struggle with issues of robustness, failing catastrophically when the underlying system dynamics are perturbed. The Robust RL formulation tackles this by adding worst-case adversarial noise to the dynamics and constructing the noise distribution as the solution to a zero-sum minimax game. However, existing work on learning solutions to the Robust RL formulation has primarily focused on training a single RL agent against a single adversary. In this work, we demonstrate that using a single adversary does not consistently yield robustness to dynamics variations under standard parametrizations of the adversary; the resulting policy is highly exploitable by new adversaries. We propose a population-based augmentation to the Robust RL formulation in which we randomly initialize a population of adversaries and sample from the population uniformly during training. We empirically validate across robotics benchmarks that the use of an adversarial population results in a more robust policy that also improves out-of-distribution generalization. Finally, we demonstrate that this approach provides comparable robustness and generalization as domain randomization on these benchmarks while avoiding a ubiquitous domain randomization failure mode.

Yuqing Du, Stas Tiomkin, Emre Kiciman et al.

One difficulty in using artificial agents for human-assistive applications lies in the challenge of accurately assisting with a person's goal(s). Existing methods tend to rely on inferring the human's goal, which is challenging when there are many potential goals or when the set of candidate goals is difficult to identify. We propose a new paradigm for assistance by instead increasing the human's ability to control their environment, and formalize this approach by augmenting reinforcement learning with human empowerment. This task-agnostic objective preserves the person's autonomy and ability to achieve any eventual state. We test our approach against assistance based on goal inference, highlighting scenarios where our method overcomes failure modes stemming from goal ambiguity or misspecification. As existing methods for estimating empowerment in continuous domains are computationally hard, precluding its use in real time learned assistance, we also propose an efficient empowerment-inspired proxy metric. Using this, we are able to successfully demonstrate our method in a shared autonomy user study for a challenging simulated teleoperation task with human-in-the-loop training.

Yuqing Du, Sheng Yang, Kaibin Huang

Edge machine learning involves the deployment of learning algorithms at the wireless network edge so as to leverage massive mobile data for enabling intelligent applications. The mainstream edge learning approach, federated learning, has been developed based on distributed gradient descent. Based on the approach, stochastic gradients are computed at edge devices and then transmitted to an edge server for updating a global AI model. Since each stochastic gradient is typically high-dimensional (with millions to billions of coefficients), communication overhead becomes a bottleneck for edge learning. To address this issue, we propose in this work a novel framework of hierarchical stochastic gradient quantization and study its effect on the learning performance. First, the framework features a practical hierarchical architecture for decomposing the stochastic gradient into its norm and normalized block gradients, and efficiently quantizes them using a uniform quantizer and a low-dimensional codebook on a Grassmann manifold, respectively. Subsequently, the quantized normalized block gradients are scaled and cascaded to yield the quantized normalized stochastic gradient using a so-called hinge vector designed under the criterion of minimum distortion. The hinge vector is also efficiently compressed using another low-dimensional Grassmannian quantizer. The other feature of the framework is a bit-allocation scheme for reducing the quantization error. The scheme determines the resolutions of the low-dimensional quantizers in the proposed framework. The framework is proved to guarantee model convergency by analyzing the convergence rate as a function of the quantization bits. Furthermore, by simulation, our design is shown to substantially reduce the communication overhead compared with the state-of-the-art signSGD scheme, while both achieve similar learning accuracies.

Qunsong Zeng, Yuqing Du, Kin K. Leung et al.

Edge machine learning involves the development of learning algorithms at the network edge to leverage massive distributed data and computation resources. Among others, the framework of federated edge learning (FEEL) is particularly promising for its data-privacy preservation. FEEL coordinates global model training at a server and local model training at edge devices over wireless links. In this work, we explore the new direction of energy-efficient radio resource management (RRM) for FEEL. To reduce devices' energy consumption, we propose energy-efficient strategies for bandwidth allocation and scheduling. They adapt to devices' channel states and computation capacities so as to reduce their sum energy consumption while warranting learning performance. In contrast with the traditional rate-maximization designs, the derived optimal policies allocate more bandwidth to those scheduled devices with weaker channels or poorer computation capacities, which are the bottlenecks of synchronized model updates in FEEL. On the other hand, the scheduling priority function derived in closed form gives preferences to devices with better channels and computation capacities. Substantial energy reduction contributed by the proposed strategies is demonstrated in learning experiments.

Guangxu Zhu, Dongzhu Liu, Yuqing Du et al.

The recent revival of artificial intelligence (AI) is revolutionizing almost every branch of science and technology. Given the ubiquitous smart mobile gadgets and Internet of Things (IoT) devices, it is expected that a majority of intelligent applications will be deployed at the edge of wireless networks. This trend has generated strong interests in realizing an "intelligent edge" to support AI-enabled applications at various edge devices. Accordingly, a new research area, called edge learning, emerges, which crosses and revolutionizes two disciplines: wireless communication and machine learning. A major theme in edge learning is to overcome the limited computing power, as well as limited data, at each edge device. This is accomplished by leveraging the mobile edge computing (MEC) platform and exploiting the massive data distributed over a large number of edge devices. In such systems, learning from distributed data and communicating between the edge server and devices are two critical and coupled aspects, and their fusion poses many new research challenges. This article advocates a new set of design principles for wireless communication in edge learning, collectively called learning-driven communication. Illustrative examples are provided to demonstrate the effectiveness of these design principles, and unique research opportunities are identified.