In the same way that the computer vision (CV) and natural language processing (NLP) communities have developed self-supervised methods, reinforcement learning (RL) can be cast as a self-supervised problem: learning to reach any goal, without requiring human-specified rewards or labels. However, actually building a self-supervised foundation for RL faces some important challenges. Building on prior contrastive approaches to this RL problem, we conduct careful ablation experiments and discover that a shallow and wide architecture, combined with careful weight initialization and data augmentation, can significantly boost the performance of these contrastive RL approaches on challenging simulated benchmarks. Additionally, we demonstrate that, with these design decisions, contrastive approaches can solve real-world robotic manipulation tasks, with tasks being specified by a single goal image provided after training.
We propose a method to fuse frozen text-only large language models (LLMs) with pre-trained image encoder and decoder models, by mapping between their embedding spaces. Our model demonstrates a wide suite of multimodal capabilities: image retrieval, novel image generation, and multimodal dialogue. Ours is the first approach capable of conditioning on arbitrarily interleaved image and text inputs to generate coherent image (and text) outputs. To achieve strong performance on image generation, we propose an efficient mapping network to ground the LLM to an off-the-shelf text-to-image generation model. This mapping network translates hidden representations of text into the embedding space of the visual models, enabling us to leverage the strong text representations of the LLM for visual outputs. Our approach outperforms baseline generation models on tasks with longer and more complex language. In addition to novel image generation, our model is also capable of image retrieval from a prespecified dataset, and decides whether to retrieve or generate at inference time. This is done with a learnt decision module which conditions on the hidden representations of the LLM. Our model exhibits a wider range of capabilities compared to prior multimodal language models. It can process image-and-text inputs, and produce retrieved images, generated images, and generated text -- outperforming non-LLM based generation models across several text-to-image tasks that measure context dependence.
Imitation learning is a powerful tool for training robot manipulation policies, allowing them to learn from expert demonstrations without manual programming or trial-and-error. However, common methods of data collection, such as human supervision, scale poorly, as they are time-consuming and labor-intensive. In contrast, Task and Motion Planning (TAMP) can autonomously generate large-scale datasets of diverse demonstrations. In this work, we show that the combination of large-scale datasets generated by TAMP supervisors and flexible Transformer models to fit them is a powerful paradigm for robot manipulation. To that end, we present a novel imitation learning system called OPTIMUS that trains large-scale visuomotor Transformer policies by imitating a TAMP agent. OPTIMUS introduces a pipeline for generating TAMP data that is specifically curated for imitation learning and can be used to train performant transformer-based policies. In this paper, we present a thorough study of the design decisions required to imitate TAMP and demonstrate that OPTIMUS can solve a wide variety of challenging vision-based manipulation tasks with over 70 different objects, ranging from long-horizon pick-and-place tasks, to shelf and articulated object manipulation, achieving 70 to 80% success rates. Video results at https://mihdalal.github.io/optimus/
Open-world survival games pose significant challenges for AI algorithms due to their multi-tasking, deep exploration, and goal prioritization requirements. Despite reinforcement learning (RL) being popular for solving games, its high sample complexity limits its effectiveness in complex open-world games like Crafter or Minecraft. We propose a novel approach, SPRING, to read the game's original academic paper and use the knowledge learned to reason and play the game through a large language model (LLM). Prompted with the LaTeX source as game context and a description of the agent's current observation, our SPRING framework employs a directed acyclic graph (DAG) with game-related questions as nodes and dependencies as edges. We identify the optimal action to take in the environment by traversing the DAG and calculating LLM responses for each node in topological order, with the LLM's answer to final node directly translating to environment actions. In our experiments, we study the quality of in-context "reasoning" induced by different forms of prompts under the setting of the Crafter open-world environment. Our experiments suggest that LLMs, when prompted with consistent chain-of-thought, have great potential in completing sophisticated high-level trajectories. Quantitatively, SPRING with GPT-4 outperforms all state-of-the-art RL baselines, trained for 1M steps, without any training. Finally, we show the potential of games as a test bed for LLMs.
Pre-trained large language models (LLMs) capture procedural knowledge about the world. Recent work has leveraged LLM's ability to generate abstract plans to simplify challenging control tasks, either by action scoring, or action modeling (fine-tuning). However, the transformer architecture inherits several constraints that make it difficult for the LLM to directly serve as the agent: e.g. limited input lengths, fine-tuning inefficiency, bias from pre-training, and incompatibility with non-text environments. To maintain compatibility with a low-level trainable actor, we propose to instead use the knowledge in LLMs to simplify the control problem, rather than solving it. We propose the Plan, Eliminate, and Track (PET) framework. The Plan module translates a task description into a list of high-level sub-tasks. The Eliminate module masks out irrelevant objects and receptacles from the observation for the current sub-task. Finally, the Track module determines whether the agent has accomplished each sub-task. On the AlfWorld instruction following benchmark, the PET framework leads to a significant 15% improvement over SOTA for generalization to human goal specifications.
The recent explosion of interest in multimodal applications has resulted in a wide selection of datasets and methods for representing and integrating information from different signals. Despite these empirical advances, there remain fundamental research questions: how can we quantify the nature of interactions that exist among input features? Subsequently, how can we capture these interactions using suitable data-driven methods? To answer this question, we propose an information-theoretic approach to quantify the degree of redundancy, uniqueness, and synergy across input features, which we term the PID statistics of a multimodal distribution. Using 2 newly proposed estimators that scale to high-dimensional distributions, we demonstrate their usefulness in quantifying the interactions within multimodal datasets, the nature of interactions captured by multimodal models, and principled approaches for model selection. We conduct extensive experiments on both synthetic datasets where the PID statistics are known and on large-scale multimodal benchmarks where PID estimation was previously impossible. Finally, to demonstrate the real-world applicability of our approach, we present three case studies in pathology, mood prediction, and robotic perception where our framework accurately recommends strong multimodal models for each application.
Data augmentation is one of the most prevalent tools in deep learning, underpinning many recent advances, including those from classification, generative models, and representation learning. The standard approach to data augmentation combines simple transformations like rotations and flips to generate new images from existing ones. However, these new images lack diversity along key semantic axes present in the data. Consider the task of recognizing different animals. Current augmentations fail to produce diversity in task-relevant high-level semantic attributes like the species of the animal. We address the lack of diversity in data augmentation with image-to-image transformations parameterized by pre-trained text-to-image diffusion models. Our method edits images to change their semantics using an off-the-shelf diffusion model, and generalizes to novel visual concepts from a few labelled examples. We evaluate our approach on image classification tasks in a few-shot setting, and on a real-world weed recognition task, and observe an improvement in accuracy in tested domains.
We propose an efficient method to ground pretrained text-only language models to the visual domain, enabling them to process and generate arbitrarily interleaved image-and-text data. Our method leverages the abilities of language models learnt from large scale text-only pretraining, such as in-context learning and free-form text generation. We keep the language model frozen, and finetune input and output linear layers to enable cross-modality interactions. This allows our model to process arbitrarily interleaved image-and-text inputs, and generate free-form text interleaved with retrieved images. We achieve strong zero-shot performance on grounded tasks such as contextual image retrieval and multimodal dialogue, and showcase compelling interactive abilities. Our approach works with any off-the-shelf language model and paves the way towards an effective, general solution for leveraging pretrained language models in visually grounded settings.
Despite recent progress towards scaling up multimodal vision-language models, these models are still known to struggle on compositional generalization benchmarks such as Winoground. We find that a critical component lacking from current vision-language models is relation-level alignment: the ability to match directional semantic relations in text (e.g., "mug in grass") with spatial relationships in the image (e.g., the position of the mug relative to the grass). To tackle this problem, we show that relation alignment can be enforced by encouraging the directed language attention from 'mug' to 'grass' (capturing the semantic relation 'in') to match the directed visual attention from the mug to the grass. Tokens and their corresponding objects are softly identified using the cross-modal attention. We prove that this notion of soft relation alignment is equivalent to enforcing congruence between vision and language attention matrices under a 'change of basis' provided by the cross-modal attention matrix. Intuitively, our approach projects visual attention into the language attention space to calculate its divergence from the actual language attention, and vice versa. We apply our Cross-modal Attention Congruence Regularization (CACR) loss to UNITER and improve on the state-of-the-art approach to Winoground.