Vision-Language-Action (VLA) models aim to predict robotic actions based on visual observations and language instructions. Existing approaches require fine-tuning pre-trained visionlanguage models (VLMs) as visual and language features are independently fed into downstream policies, degrading the pre-trained semantic alignments. We propose OTTER, a novel VLA architecture that leverages these existing alignments through explicit, text-aware visual feature extraction. Instead of processing all visual features, OTTER selectively extracts and passes only task-relevant visual features that are semantically aligned with the language instruction to the policy transformer. This allows OTTER to keep the pre-trained vision-language encoders frozen. Thereby, OTTER preserves and utilizes the rich semantic understanding learned from large-scale pre-training, enabling strong zero-shot generalization capabilities. In simulation and real-world experiments, OTTER significantly outperforms existing VLA models, demonstrating strong zeroshot generalization to novel objects and environments. Video, code, checkpoints, and dataset: https://ottervla.github.io/.

Simulation offers a promising approach for cheaply scaling training data for generalist policies. To scalably generate data from diverse and realistic tasks, existing algorithms either rely on large language models (LLMs) that may hallucinate tasks not interesting for robotics; or digital twins, which require careful real-to-sim alignment and are hard to scale. To address these challenges, we introduce Video2Policy, a novel framework that leverages internet RGB videos to reconstruct tasks based on everyday human behavior. Our approach comprises two phases: (1) task generation in simulation from videos; and (2) reinforcement learning utilizing in-context LLM-generated reward functions iteratively. We demonstrate the efficacy of Video2Policy by reconstructing over 100 videos from the Something-Something-v2 (SSv2) dataset, which depicts diverse and complex human behaviors on 9 different tasks. Our method can successfully train RL policies on such tasks, including complex and challenging tasks such as throwing. Finally, we show that the generated simulation data can be scaled up for training a general policy, and it can be transferred back to the real robot in a Real2Sim2Real way.

This paper enables real-world humanoid robots to maintain stability while performing expressive motions like humans do. We propose ExBody2, a generalized whole-body tracking framework that can take any reference motion inputs and control the humanoid to mimic the motion. The model is trained in simulation with Reinforcement Learning and then transferred to the real world. It decouples keypoint tracking with velocity control, and effectively leverages a privileged teacher policy to distill precise mimic skills into the target student policy, which enables high-fidelity replication of dynamic movements such as running, crouching, dancing, and other challenging motions. We present a comprehensive qualitative and quantitative analysis of crucial design factors in the paper. We conduct our experiments on two humanoid platforms and demonstrate the superiority of our approach against state-of-the-arts, providing practical guidelines to pursue the extreme of whole-body control for humanoid robots.

Open-vocabulary generalization requires robotic systems to perform tasks involving complex and diverse environments and task goals. While the recent advances in vision language models (VLMs) present unprecedented opportunities to solve unseen problems, how to utilize their emergent capabilities to control robots in the physical world remains an open question. In this paper, we present MOKA (Marking Open-vocabulary Keypoint Affordances), an approach that employs VLMs to solve robotic manipulation tasks specified by free-form language descriptions. At the heart of our approach is a compact point-based representation of affordance and motion that bridges the VLM's predictions on RGB images and the robot's motions in the physical world. By prompting a VLM pre-trained on Internet-scale data, our approach predicts the affordances and generates the corresponding motions by leveraging the concept understanding and commonsense knowledge from broad sources. To scaffold the VLM's reasoning in zero-shot, we propose a visual prompting technique that annotates marks on the images, converting the prediction of keypoints and waypoints into a series of visual question answering problems that are feasible for the VLM to solve. Using the robot experiences collected in this way, we further investigate ways to bootstrap the performance through in-context learning and policy distillation. We evaluate and analyze MOKA's performance on a variety of manipulation tasks specified by free-form language descriptions, such as tool use, deformable body manipulation, and object rearrangement.

In this paper, we propose a real-world benchmark for studying robotic learning in the context of functional manipulation: a robot needs to accomplish complex long-horizon behaviors by composing individual manipulation skills in functionally relevant ways. The core design principles of our Functional Manipulation Benchmark (FMB) emphasize a harmonious balance between complexity and accessibility. Tasks are deliberately scoped to be narrow, ensuring that models and datasets of manageable scale can be utilized effectively to track progress. Simultaneously, they are diverse enough to pose a significant generalization challenge. Furthermore, the benchmark is designed to be easily replicable, encompassing all essential hardware and software components. To achieve this goal, FMB consists of a variety of 3D-printed objects designed for easy and accurate replication by other researchers. The objects are procedurally generated, providing a principled framework to study generalization in a controlled fashion. We focus on fundamental manipulation skills, including grasping, repositioning, and a range of assembly behaviors. The FMB can be used to evaluate methods for acquiring individual skills, as well as methods for combining and ordering such skills to solve complex, multi-stage manipulation tasks. We also offer an imitation learning framework that includes a suite of policies trained to solve the proposed tasks. This enables researchers to utilize our tasks as a versatile toolkit for examining various parts of the pipeline. For example, researchers could propose a better design for a grasping controller and evaluate it in combination with our baseline reorientation and assembly policies as part of a pipeline for solving multi-stage tasks. Our dataset, object CAD files, code, and evaluation videos can be found on our project website: https://functional-manipulation-benchmark.github.io

The existing internet-scale image and video datasets cover a wide range of everyday objects and tasks, bringing the potential of learning policies that generalize in diverse scenarios. Prior works have explored visual pre-training with different self-supervised objectives. Still, the generalization capabilities of the learned policies and the advantages over well-tuned baselines remain unclear from prior studies. In this work, we present a focused study of the generalization capabilities of the pre-trained visual representations at the categorical level. We identify the key bottleneck in using a frozen pre-trained visual backbone for policy learning and then propose SpawnNet, a novel two-stream architecture that learns to fuse pre-trained multi-layer representations into a separate network to learn a robust policy. Through extensive simulated and real experiments, we show significantly better categorical generalization compared to prior approaches in imitation learning settings. Open-sourced code and videos can be found on our website: https://xingyu-lin.github.io/spawnnet.

We are interested in learning scalable agents for reinforcement learning that can learn from large-scale, diverse sequential data similar to current large vision and language models. To this end, this paper presents masked decision prediction (MaskDP), a simple and scalable self-supervised pretraining method for reinforcement learning (RL) and behavioral cloning (BC). In our MaskDP approach, we employ a masked autoencoder (MAE) to state-action trajectories, wherein we randomly mask state and action tokens and reconstruct the missing data. By doing so, the model is required to infer masked-out states and actions and extract information about dynamics. We find that masking different proportions of the input sequence significantly helps with learning a better model that generalizes well to multiple downstream tasks. In our empirical study, we find that a MaskDP model gains the capability of zero-shot transfer to new BC tasks, such as single and multiple goal reaching, and it can zero-shot infer skills from a few example transitions. In addition, MaskDP transfers well to offline RL and shows promising scaling behavior w.r.t. to model size. It is amenable to data-efficient finetuning, achieving competitive results with prior methods based on autoregressive pretraining.

Visual model-based reinforcement learning (RL) has the potential to enable sample-efficient robot learning from visual observations. Yet the current approaches typically train a single model end-to-end for learning both visual representations and dynamics, making it difficult to accurately model the interaction between robots and small objects. In this work, we introduce a visual model-based RL framework that decouples visual representation learning and dynamics learning. Specifically, we train an autoencoder with convolutional layers and vision transformers (ViT) to reconstruct pixels given masked convolutional features, and learn a latent dynamics model that operates on the representations from the autoencoder. Moreover, to encode task-relevant information, we introduce an auxiliary reward prediction objective for the autoencoder. We continually update both autoencoder and dynamics model using online samples collected from environment interaction. We demonstrate that our decoupling approach achieves state-of-the-art performance on a variety of visual robotic tasks from Meta-world and RLBench, e.g., we achieve 81.7% success rate on 50 visual robotic manipulation tasks from Meta-world, while the baseline achieves 67.9%. Code is available on the project website: https://sites.google.com/view/mwm-rl.

We study generalizable policy learning from demonstrations for complex low-level control tasks (e.g., contact-rich object manipulations). We propose an imitation learning method that incorporates the idea of temporal abstraction and the planning capabilities from Hierarchical RL (HRL) in a novel and effective manner. As a step towards decision foundation models, our design can utilize scalable, albeit highly sub-optimal, demonstrations. Specifically, we find certain short subsequences of the demos, i.e. the chain-of-thought (CoT), reflect their hierarchical structures by marking the completion of subgoals in the tasks. Our model learns to dynamically predict the entire CoT as coherent and structured long-term action guidance and consistently outperforms typical two-stage subgoal-conditioned policies. On the other hand, such CoT facilitates generalizable policy learning as they exemplify the decision patterns shared among demos (even those with heavy noises and randomness). Our method, Chain-of-Thought Predictive Control (CoTPC), significantly outperforms existing ones on challenging low-level manipulation tasks from scalable yet highly sub-optimal demos.

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.