Learning a generalist embodied agent capable of completing multiple tasks poses challenges, primarily stemming from the scarcity of action-labeled robotic datasets. In contrast, a vast amount of human videos exist, capturing intricate tasks and interactions with the physical world. Promising prospects arise for utilizing actionless human videos for pre-training and transferring the knowledge to facilitate robot policy learning through limited robot demonstrations. In this paper, we introduce a novel framework that leverages a unified discrete diffusion to combine generative pre-training on human videos and policy fine-tuning on a small number of action-labeled robot videos. We start by compressing both human and robot videos into unified video tokens. In the pre-training stage, we employ a discrete diffusion model with a mask-and-replace diffusion strategy to predict future video tokens in the latent space. In the fine-tuning stage, we harness the imagined future videos to guide low-level action learning trained on a limited set of robot data. Experiments demonstrate that our method generates high-fidelity future videos for planning and enhances the fine-tuned policies compared to previous state-of-the-art approaches with superior generalization ability. Our project website is available at https://video-diff.github.io/.

Bimanual handovers are crucial for transferring large, deformable or delicate objects. This paper proposes a framework for generating kinematically constrained human-like bimanual robot motions to ensure seamless and natural robot-to-human object handovers. We use a Hidden Semi-Markov Model (HSMM) to reactively generate suitable response trajectories for a robot based on the observed human partner's motion. The trajectories are adapted with task space constraints to ensure accurate handovers. Results from a pilot study show that our approach is perceived as more human--like compared to a baseline Inverse Kinematics approach.

We study off-policy evaluation (OPE) in partially observable environments with complex observations, with the goal of developing estimators whose guarantee avoids exponential dependence on the horizon. While such estimators exist for MDPs and POMDPs can be converted to history-based MDPs, their estimation errors depend on the state-density ratio for MDPs which becomes history ratios after conversion, an exponential object. Recently, Uehara et al. (2022) proposed future-dependent value functions as a promising framework to address this issue, where the guarantee for memoryless policies depends on the density ratio over the latent state space. However, it also depends on the boundedness of the future-dependent value function and other related quantities, which we show could be exponential-in-length and thus erasing the advantage of the method. In this paper, we discover novel coverage assumptions tailored to the structure of POMDPs, such as outcome coverage and belief coverage. These assumptions not only enable polynomial bounds on the aforementioned quantities, but also lead to the discovery of new algorithms with complementary properties.

The incredible success of transformers on sequence modeling tasks can be largely attributed to the self-attention mechanism, which allows information to be transferred between different parts of a sequence. Self-attention allows transformers to encode causal structure which makes them particularly suitable for sequence modeling. However, the process by which transformers learn such causal structure via gradient-based training algorithms remains poorly understood. To better understand this process, we introduce an in-context learning task that requires learning latent causal structure. We prove that gradient descent on a simplified two-layer transformer learns to solve this task by encoding the latent causal graph in the first attention layer. The key insight of our proof is that the gradient of the attention matrix encodes the mutual information between tokens. As a consequence of the data processing inequality, the largest entries of this gradient correspond to edges in the latent causal graph. As a special case, when the sequences are generated from in-context Markov chains, we prove that transformers learn an induction head (Olsson et al., 2022). We confirm our theoretical findings by showing that transformers trained on our in-context learning task are able to recover a wide variety of causal structures.

Adaptive time series forecasting is essential for prediction under regime changes. Several classical methods assume linear Gaussian state space model (LGSSM) with variances constant in time. However, there are many real-world processes that cannot be captured by such models. We consider a state-space model with Markov switching variances. Such dynamical systems are usually intractable because of their computational complexity increasing exponentially with time; Variational Bayes (VB) techniques have been applied to this problem. In this paper, we propose a new way of estimating variances based on online learning theory; we adapt expert aggregation methods to learn the variances over time. We apply the proposed method to synthetic data and to the problem of electricity load forecasting. We show that this method is robust to misspecification and outperforms traditional expert aggregation.

We propose a novel approach to the problem of controller design for environments modeled as Markov decision processes (MDPs). Specifically, we consider a hierarchical MDP a graph with each vertex populated by an MDP called a "room." We first apply deep reinforcement learning (DRL) to obtain low-level policies for each room, scaling to large rooms of unknown structure. We then apply reactive synthesis to obtain a high-level planner that chooses which low-level policy to execute in each room. The central challenge in synthesizing the planner is the need for modeling rooms. We address this challenge by developing a DRL procedure to train concise "latent" policies together with PAC guarantees on their performance. Unlike previous approaches, ours circumvents a model distillation step. Our approach combats sparse rewards in DRL and enables reusability of low-level policies. We demonstrate feasibility in a case study involving agent navigation amid moving obstacles.

The framework of the Configurable Markov Decision Processes (Conf-MDPs, Metelli et al., 2018, 2019, 2022) has been introduced in recent years to model a wide range of real-world scenarios in which an agent has the opportunity to alter some environmental parameters in order to improve its learning experience. Conf-MDPs can be thought to as an extension of the traditional Markov Decision Processes (MDP, Puterman, 1994) to account for scenarios that emerge quite often in the Reinforcement Learning (RL, Sutton and Barto, 2018) problems, in which the environment rarely represents an immutable entity and can, indeed, be subject to partial control. In the Conf-MDP framework, the activity of altering the environmental parameters is named environment configuration and serves different purposes. In the simplest scenario, the configuration is carried out by the agent itself that acts as a configurator. This might suggest, at a first sight, that environment configuration can be modeled within the agent actuation.

Artificial Intelligence (AI) holds promise as a technology that can be used to improve government and economic policy-making. This paper proposes a new research agenda towards this end by introducing Social Environment Design, a general framework for the use of AI for automated policy-making that connects with the Reinforcement Learning, EconCS, and Computational Social Choice communities. The framework seeks to capture general economic environments, includes voting on policy objectives, and gives a direction for the systematic analysis of government and economic policy through AI simulation. We highlight key open problems for future research in AI-based policy-making. By solving these challenges, we hope to achieve various social welfare objectives, thereby promoting more ethical and responsible decision making.

The proposed solver is based on the Ohzeki method that combines Markov-chain Monte-Carlo (MCMC) and gradient descent, and its step sizes are trained by minimizing a loss function. In the training process, we propose a sampling-based gradient estimation that substitutes auto-differentiation with a variance estimation, thereby circumventing the failure of back propagation due to the non-differentiability of MCMC. The numerical results for a few COPs demonstrated that the proposed solver significantly accelerated the convergence speed compared with the original Ohzeki method. Combinatorial optimization problems (COPs) comprising discrete variables are considered hard to solve exactly in polynomial time, which relates to the well-known P vs. NP problem. Along with deterministic approximation algorithms, samplers such as Markovchain Monte-Carlo (MCMC) have been applied to COPs. However, the convergence time for obtaining reasonable approximate solutions is long.

Modern language models rely on the transformer architecture and attention mechanism to perform language understanding and text generation. In this work, we study learning a 1-layer self-attention model from a set of prompts and associated output data sampled from the model. We first establish a precise mapping between the self-attention mechanism and Markov models: Inputting a prompt to the model samples the output token according to a context-conditioned Markov chain (CCMC) which weights the transition matrix of a base Markov chain. Additionally, incorporating positional encoding results in position-dependent scaling of the transition probabilities. Building on this formalism, we develop identifiability/coverage conditions for the prompt distribution that guarantee consistent estimation and establish sample complexity guarantees under IID samples. Finally, we study the problem of learning from a single output trajectory generated from an initial prompt. We characterize an intriguing winner-takes-all phenomenon where the generative process implemented by self-attention collapses into sampling a limited subset of tokens due to its non-mixing nature. This provides a mathematical explanation to the tendency of modern LLMs to generate repetitive text. In summary, the equivalence to CCMC provides a simple but powerful framework to study self-attention and its properties.