Multimodal Masked Autoencoders Learn Transferable Representations

Building scalable models to learn from diverse, multimodal data remains an open challenge. For vision-language data, the dominant approaches are based on contrastive learning objectives that train a separate encoder for each modality. While effective, contrastive learning approaches introduce sampling bias depending on the data augmentations used, which can degrade performance on downstream tasks. Moreover, these methods are limited to paired image-text data, and cannot leverage widely-available unpaired data. In this paper, we investigate whether a large multimodal model trained purely via masked token prediction, without using modality-specific encoders or contrastive learning, can learn transferable representations for downstream tasks. We propose a simple and scalable network architecture, the Multimodal Masked Autoencoder (M3AE), which learns a unified encoder for both vision and language data via masked token prediction. We provide an empirical study of M3AE trained on a large-scale image-text dataset, and find that M3AE is able to learn generalizable representations that transfer well to downstream tasks. Surprisingly, we find that M3AE benefits from a higher text mask ratio (50-90%), in contrast to BERT whose standard masking ratio is 15%, due to the joint training of two data modalities. We also provide qualitative analysis showing that the learned representation incorporates meaningful information from both image and language. Lastly, we demonstrate the scalability of M3AE with larger model size and training time, and its flexibility to train on both paired image-text data as well as unpaired data.

Reward Uncertainty for Exploration in Preference-based Reinforcement Learning

Conveying complex objectives to reinforcement learning (RL) agents often requires meticulous reward engineering. Preference-based RL methods are able to learn a more flexible reward model based on human preferences by actively incorporating human feedback, i.e. teacher's preferences between two clips of behaviors. However, poor feedback-efficiency still remains a problem in current preference-based RL algorithms, as tailored human feedback is very expensive. To handle this issue, previous methods have mainly focused on improving query selection and policy initialization. At the same time, recent exploration methods have proven to be a recipe for improving sample-efficiency in RL. We present an exploration method specifically for preference-based RL algorithms. Our main idea is to design an intrinsic reward by measuring the novelty based on learned reward. Specifically, we utilize disagreement across ensemble of learned reward models. Our intuition is that disagreement in learned reward model reflects uncertainty in tailored human feedback and could be useful for exploration. Our experiments show that exploration bonus from uncertainty in learned reward improves both feedback- and sample-efficiency of preference-based RL algorithms on complex robot manipulation tasks from MetaWorld benchmarks, compared with other existing exploration methods that measure the novelty of state visitation.

Chain of Thought Imitation with Procedure Cloning

Imitation learning aims to extract high-performance policies from logged demonstrations of expert behavior. It is common to frame imitation learning as a supervised learning problem in which one fits a function approximator to the input-output mapping exhibited by the logged demonstrations (input observations to output actions). While the framing of imitation learning as a supervised input-output learning problem allows for applicability in a wide variety of settings, it is also an overly simplistic view of the problem in situations where the expert demonstrations provide much richer insight into expert behavior. For example, applications such as path navigation, robot manipulation, and strategy games acquire expert demonstrations via planning, search, or some other multi-step algorithm, revealing not just the output action to be imitated but also the procedure for how to determine this action. While these intermediate computations may use tools not available to the agent during inference (e.g., environment simulators), they are nevertheless informative as a way to explain an expert's mapping of state to actions. To properly leverage expert procedure information without relying on the privileged tools the expert may have used to perform the procedure, we propose procedure cloning, which applies supervised sequence prediction to imitate the series of expert computations. This way, procedure cloning learns not only what to do (i.e., the output action), but how and why to do it (i.e., the procedure). Through empirical analysis on navigation, simulated robotic manipulation, and game-playing environments, we show that imitating the intermediate computations of an expert's behavior enables procedure cloning to learn policies exhibiting significant generalization to unseen environment configurations, including those configurations for which running the expert's procedure directly is infeasible.

An Empirical Investigation of Representation Learning for Imitation

Imitation learning often needs a large demonstration set in order to handle the full range of situations that an agent might find itself in during deployment. However, collecting expert demonstrations can be expensive. Recent work in vision, reinforcement learning, and NLP has shown that auxiliary representation learning objectives can reduce the need for large amounts of expensive, task-specific data. Our Empirical Investigation of Representation Learning for Imitation (EIRLI) investigates whether similar benefits apply to imitation learning. We propose a modular framework for constructing representation learning algorithms, then use our framework to evaluate the utility of representation learning for imitation across several environment suites. In the settings we evaluate, we find that existing algorithms for image-based representation learning provide limited value relative to a well-tuned baseline with image augmentations. To explain this result, we investigate differences between imitation learning and other settings where representation learning has provided significant benefit, such as image classification. Finally, we release a well-documented codebase which both replicates our findings and provides a modular framework for creating new representation learning algorithms out of reusable components.

Coarse-to-fine Q-attention with Tree Expansion

Coarse-to-fine Q-attention enables sample-efficient robot manipulation by discretizing the translation space in a coarse-to-fine manner, where the resolution gradually increases at each layer in the hierarchy. Although effective, Q-attention suffers from "coarse ambiguity" - when voxelization is significantly coarse, it is not feasible to distinguish similar-looking objects without first inspecting at a finer resolution. To combat this, we propose to envision Q-attention as a tree that can be expanded and used to accumulate value estimates across the top-k voxels at each Q-attention depth. When our extension, Q-attention with Tree Expansion (QTE), replaces standard Q-attention in the Attention-driven Robot Manipulation (ARM) system, we are able to accomplish a larger set of tasks; especially on those that suffer from "coarse ambiguity". In addition to evaluating our approach across 12 RLBench tasks, we also show that the improved performance is visible in a real-world task involving small objects.

Imitating, Fast and Slow: Robust learning from demonstrations via decision-time planning

The goal of imitation learning is to mimic expert behavior from demonstrations, without access to an explicit reward signal. A popular class of approach infers the (unknown) reward function via inverse reinforcement learning (IRL) followed by maximizing this reward function via reinforcement learning (RL). The policies learned via these approaches are however very brittle in practice and deteriorate quickly even with small test-time perturbations due to compounding errors. We propose Imitation with Planning at Test-time (IMPLANT), a new meta-algorithm for imitation learning that utilizes decision-time planning to correct for compounding errors of any base imitation policy. In contrast to existing approaches, we retain both the imitation policy and the rewards model at decision-time, thereby benefiting from the learning signal of the two components. Empirically, we demonstrate that IMPLANT significantly outperforms benchmark imitation learning approaches on standard control environments and excels at zero-shot generalization when subject to challenging perturbations in test-time dynamics.

Sim-to-Real 6D Object Pose Estimation via Iterative Self-training for Robotic Bin-picking

In this paper, we propose an iterative self-training framework for sim-to-real 6D object pose estimation to facilitate cost-effective robotic grasping. Given a bin-picking scenario, we establish a photo-realistic simulator to synthesize abundant virtual data, and use this to train an initial pose estimation network. This network then takes the role of a teacher model, which generates pose predictions for unlabeled real data. With these predictions, we further design a comprehensive adaptive selection scheme to distinguish reliable results, and leverage them as pseudo labels to update a student model for pose estimation on real data. To continuously improve the quality of pseudo labels, we iterate the above steps by taking the trained student model as a new teacher and re-label real data using the refined teacher model. We evaluate our method on a public benchmark and our newly-released dataset, achieving an ADD(-S) improvement of 11.49% and 22.62% respectively. Our method is also able to improve robotic bin-picking success by 19.54%, demonstrating the potential of iterative sim-to-real solutions for robotic applications.

Coarse-to-Fine Q-attention with Learned Path Ranking

We propose Learned Path Ranking (LPR), a method that accepts an end-effector goal pose, and learns to rank a set of goal-reaching paths generated from an array of path generating methods, including: path planning, Bezier curve sampling, and a learned policy. The core idea being that each of the path generation modules will be useful in different tasks, or at different stages in a task. When LPR is added as an extension to C2F-ARM, our new system, C2F-ARM+LPR, retains the sample efficiency of its predecessor, while also being able to accomplish a larger set of tasks; in particular, tasks that require very specific motions (e.g. opening toilet seat) that need to be inferred from both demonstrations and exploration data. In addition to benchmarking our approach across 16 RLBench tasks, we also learn real-world tasks, tabula rasa, in 10-15 minutes, with only 3 demonstrations.