Foundation models pretrained on diverse data at scale have demonstrated extraordinary capabilities in a wide range of vision and language tasks. When such models are deployed in real world environments, they inevitably interface with other entities and agents. For example, language models are often used to interact with human beings through dialogue, and visual perception models are used to autonomously navigate neighborhood streets. In response to these developments, new paradigms are emerging for training foundation models to interact with other agents and perform long-term reasoning. These paradigms leverage the existence of ever-larger datasets curated for multimodal, multitask, and generalist interaction. Research at the intersection of foundation models and decision making holds tremendous promise for creating powerful new systems that can interact effectively across a diverse range of applications such as dialogue, autonomous driving, healthcare, education, and robotics. In this manuscript, we examine the scope of foundation models for decision making, and provide conceptual tools and technical background for understanding the problem space and exploring new research directions. We review recent approaches that ground foundation models in practical decision making applications through a variety of methods such as prompting, conditional generative modeling, planning, optimal control, and reinforcement learning, and discuss common challenges and open problems in the field.
Preference-based reinforcement learning (RL) provides a framework to train agents using human preferences between two behaviors. However, preference-based RL has been challenging to scale since it requires a large amount of human feedback to learn a reward function aligned with human intent. In this paper, we present Preference Transformer, a neural architecture that models human preferences using transformers. Unlike prior approaches assuming human judgment is based on the Markovian rewards which contribute to the decision equally, we introduce a new preference model based on the weighted sum of non-Markovian rewards. We then design the proposed preference model using a transformer architecture that stacks causal and bidirectional self-attention layers. We demonstrate that Preference Transformer can solve a variety of control tasks using real human preferences, while prior approaches fail to work. We also show that Preference Transformer can induce a well-specified reward and attend to critical events in the trajectory by automatically capturing the temporal dependencies in human decision-making. Code is available on the project website: https://sites.google.com/view/preference-transformer.
Learning from human preferences is important for language models to be helpful and useful for humans, and to align with human and social values. Prior work have achieved remarkable successes by learning from human feedback to understand and follow instructions. Nonetheless, these methods are either founded on hand-picked model generations that are favored by human annotators, rendering them ineffective in terms of data utilization and challenging to apply in general, or they depend on reward functions and reinforcement learning, which are prone to imperfect reward function and extremely challenging to optimize. In this work, we propose a novel technique, Chain of Hindsight, that is easy to optimize and can learn from any form of feedback, regardless of its polarity. Our idea is inspired by how humans learn from extensive feedback presented in the form of languages. We convert all types of feedback into sentences, which are then used to fine-tune the model, allowing us to take advantage of the language comprehension capabilities of language models. We condition the model on a sequence of model generations paired with feedback. By doing so, models are trained to generate outputs based on feedback, and models can learn to identify and correct negative attributes or errors. Applying our method to large language models, we observed that Chain of Hindsight significantly surpasses previous methods in aligning language models with human preferences. We observed significant improvements on summarization and dialogue tasks and our approach is markedly preferred in human evaluations.
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.
This work aims to push the limits of agility for bipedal robots by enabling a torque-controlled bipedal robot to perform robust and versatile dynamic jumps in the real world. We present a multi-task reinforcement learning framework to train the robot to accomplish a large variety of jumping tasks, such as jumping to different locations and directions. To improve performance on these challenging tasks, we develop a new policy structure that encodes the robot's long-term input/output (I/O) history while also providing direct access to its short-term I/O history. In order to train a versatile multi-task policy, we utilize a multi-stage training scheme that includes different training stages for different objectives. After multi-stage training, the multi-task policy can be directly transferred to Cassie, a physical bipedal robot. Training on different tasks and exploring more diverse scenarios leads to highly robust policies that can exploit the diverse set of learned skills to recover from perturbations or poor landings during real-world deployment. Such robustness in the proposed multi-task policy enables Cassie to succeed in completing a variety of challenging jump tasks in the real world, such as standing long jumps, jumping onto elevated platforms, and multi-axis jumps.
Learning from human preferences is important for language models to be helpful and useful for humans, and to align with human and social values. Existing works focus on supervised finetuning of pretrained models, based on curated model generations that are preferred by human labelers. Such works have achieved remarkable successes in understanding and following instructions (e.g., InstructGPT, ChatGPT, etc). However, to date, a key limitation of supervised finetuning is that it cannot learn from negative ratings; models are only trained on positive-rated data, which makes it data inefficient. Because collecting human feedback data is both time consuming and expensive, it is vital for the model to learn from all feedback, akin to the remarkable ability of humans to learn from diverse feedback. In this work, we propose a novel technique called Hindsight Finetuning for making language models learn from diverse human feedback. In fact, our idea is motivated by how humans learn from hindsight experience. We condition the model on a sequence of model generations paired with hindsight feedback, and finetune the model to predict the most preferred output. By doing so, models can learn to identify and correct negative attributes or errors. Applying the method to GPT-J, we observe that it significantly improves results on summarization and dialogue tasks using the same amount of human feedback.
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.
One of the key capabilities of intelligent agents is the ability to discover useful skills without external supervision. However, the current unsupervised skill discovery methods are often limited to acquiring simple, easy-to-learn skills due to the lack of incentives to discover more complex, challenging behaviors. We introduce a novel unsupervised skill discovery method, Controllability-aware Skill Discovery (CSD), which actively seeks complex, hard-to-control skills without supervision. The key component of CSD is a controllability-aware distance function, which assigns larger values to state transitions that are harder to achieve with the current skills. Combined with distance-maximizing skill discovery, CSD progressively learns more challenging skills over the course of training as our jointly trained distance function reduces rewards for easy-to-achieve skills. Our experimental results in six robotic manipulation and locomotion environments demonstrate that CSD can discover diverse complex skills including object manipulation and locomotion skills with no supervision, significantly outperforming prior unsupervised skill discovery methods. Videos and code are available at https://seohong.me/projects/csd/
Reinforcement learning has seen wide success in finetuning large language models to better align with instructions via human feedback. The so-called algorithm, Reinforcement Learning with Human Feedback (RLHF) demonstrates impressive performance on the GPT series models. However, the underlying Reinforcement Learning (RL) algorithm is complex and requires an additional training pipeline for reward and value networks. In this paper, we consider an alternative approach: converting feedback to instruction by relabeling the original one and training the model for better alignment in a supervised manner. Such an algorithm doesn't require any additional parameters except for the original language model and maximally reuses the pretraining pipeline. To achieve this, we formulate instruction alignment problem for language models as a goal-reaching problem in decision making. We propose Hindsight Instruction Relabeling (HIR), a novel algorithm for aligning language models with instructions. The resulting two-stage algorithm shed light to a family of reward-free approaches that utilize the hindsightly relabeled instructions based on feedback. We evaluate the performance of HIR extensively on 12 challenging BigBench reasoning tasks and show that HIR outperforms the baseline algorithms and is comparable to or even surpasses supervised finetuning.
Learning from human preferences is important for language models to be helpful and useful for humans, and to align with human and social values. Existing works focus on supervised finetuning of pretrained models, based on curated model generations that are preferred by human labelers. Such works have achieved remarkable successes in understanding and following instructions (e.g., InstructGPT, ChatGPT, etc). However, to date, a key limitation of supervised finetuning is that it cannot learn from negative ratings; models are only trained on positive-rated data, which makes it data inefficient. Because collecting human feedback data is both time consuming and expensive, it is vital for the model to learn from all feedback, akin to the remarkable ability of humans to learn from diverse feedback. In this work, we propose a novel technique called Hindsight Finetuning for making language models learn from diverse human feedback. In fact, our idea is motivated by how humans learn from hindsight experience. We condition the model on a sequence of model generations paired with hindsight feedback, and finetune the model to predict the most preferred output. By doing so, models can learn to identify and correct negative attributes or errors. Applying the method to GPT-J, we observe that it significantly improves results on summarization and dialogue tasks using the same amount of human feedback.