Unlike traditional methods that focus on evaluating single state-action pairs or apply action chunking in the actor network, this approach feeds chunked actions directly into the critic. Leveraging the Transformer's strength in processing sequential data, the proposed architecture achieves more robust value estimation. Empirical evaluations demonstrate that this method leads to efficient and stable training, particularly excelling in environments with sparse rewards or Multi-Phase tasks.Contribution(s) 1. We present a novel critic architecture for SAC that leverages Transformers to process sequential information, resulting in more accurate value estimations. Context: Transformer-Based Critic Network 2. We introduce a method for incorporating N-Step returns into the critic network in a stable and efficient manner, effectively mitigating the common challenges of variance and importance sampling associated with N-returns. Context: Stable Integration of N-Returns 3. We shift action chunking from the actor to the critic, demonstrating that enhanced temporal reasoning at the critic level--beyond traditional actor-side exploration--drives performance improvements in sparse and multi-phase tasks. Context: Unlike previous approaches that focus on actor-side chunking for exploration, our Transformer-based critic network produces a smooth value surface that is highly responsive to dataset variations, eliminating the need for additional exploration enhancements.

We provide a general framework for learning diffusion bridges that transport prior to target distributions. It includes existing diffusion models for generative modeling, but also underdamped versions with degenerate diffusion matrices, where the noise only acts in certain dimensions. Extending previous findings, our framework allows to rigorously show that score matching in the underdamped case is indeed equivalent to maximizing a lower bound on the likelihood. Motivated by superior convergence properties and compatibility with sophisticated numerical integration schemes of underdamped stochastic processes, we propose \emph{underdamped diffusion bridges}, where a general density evolution is learned rather than prescribed by a fixed noising process. We apply our method to the challenging task of sampling from unnormalized densities without access to samples from the target distribution. Across a diverse range of sampling problems, our approach demonstrates state-of-the-art performance, notably outperforming alternative methods, while requiring significantly fewer discretization steps and no hyperparameter tuning.

Diffusion models optimized via variational inference (VI) have emerged as a promising tool for generating samples from unnormalized target densities. These models create samples by simulating a stochastic differential equation, starting from a simple, tractable prior, typically a Gaussian distribution. However, when the support of this prior differs greatly from that of the target distribution, diffusion models often struggle to explore effectively or suffer from large discretization errors. Moreover, learning the prior distribution can lead to mode-collapse, exacerbated by the mode-seeking nature of reverse Kullback-Leibler divergence commonly used in VI. To address these challenges, we propose end-to-end learnable Gaussian mixture priors (GMPs). GMPs offer improved control over exploration, adaptability to target support, and increased expressiveness to counteract mode collapse. We further leverage the structure of mixture models by proposing a strategy to iteratively refine the model by adding mixture components during training. Our experimental results demonstrate significant performance improvements across a diverse range of real-world and synthetic benchmark problems when using GMPs without requiring additional target evaluations.

This paper introduces IRIS, an immersive Robot Interaction System leveraging Extended Reality (XR), designed for robot data collection and interaction across multiple simulators, benchmarks, and real-world scenarios. While existing XR-based data collection systems provide efficient and intuitive solutions for large-scale data collection, they are often challenging to reproduce and reuse. This limitation arises because current systems are highly tailored to simulator-specific use cases and environments. IRIS is a novel, easily extendable framework that already supports multiple simulators, benchmarks, and even headsets. Furthermore, IRIS is able to include additional information from real-world sensors, such as point clouds captured through depth cameras. A unified scene specification is generated directly from simulators or real-world sensors and transmitted to XR headsets, creating identical scenes in XR. This specification allows IRIS to support any of the objects, assets, and robots provided by the simulators. In addition, IRIS introduces shared spatial anchors and a robust communication protocol that links simulations between multiple XR headsets. This feature enables multiple XR headsets to share a synchronized scene, facilitating collaborative and multi-user data collection. IRIS can be deployed on any device that supports the Unity Framework, encompassing the vast majority of commercially available headsets. In this work, IRIS was deployed and tested on the Meta Quest 3 and the HoloLens 2. IRIS showcased its versatility across a wide range of real-world and simulated scenarios, using current popular robot simulators such as MuJoCo, IsaacSim, CoppeliaSim, and Genesis. In addition, a user study evaluates IRIS on a data collection task for the LIBERO benchmark. The study shows that IRIS significantly outperforms the baseline in both objective and subjective metrics.

Designing modern imitation learning (IL) policies requires making numerous decisions, including the selection of feature encoding, architecture, policy representation, and more. As the field rapidly advances, the range of available options continues to grow, creating a vast and largely unexplored design space for IL policies. In this work, we present X-IL, an accessible open-source framework designed to systematically explore this design space. The framework's modular design enables seamless swapping of policy components, such as backbones (e.g., Transformer, Mamba, xLSTM) and policy optimization techniques (e.g., Score-matching, Flow-matching). This flexibility facilitates comprehensive experimentation and has led to the discovery of novel policy configurations that outperform existing methods on recent robot learning benchmarks. Our experiments demonstrate not only significant performance gains but also provide valuable insights into the strengths and weaknesses of various design choices. This study serves as both a practical reference for practitioners and a foundation for guiding future research in imitation learning.

Learning for manipulation requires using policies that have access to rich sensory information such as point clouds or RGB images. Point clouds efficiently capture geometric structures, making them essential for manipulation tasks in imitation learning. In contrast, RGB images provide rich texture and semantic information that can be crucial for certain tasks. Existing approaches for fusing both modalities assign 2D image features to point clouds. However, such approaches often lose global contextual information from the original images. In this work, we propose a novel imitation learning method that effectively combines the strengths of both point cloud and RGB modalities. Our method conditions the point-cloud encoder on global and local image tokens using adaptive layer norm conditioning, leveraging the beneficial properties of both modalities. Through extensive experiments on the challenging RoboCasa benchmark, we demonstrate the limitations of relying on either modality alone and show that our method achieves state-of-the-art performance across all tasks.

Manipulating objects with varying geometries and deformable objects is a major challenge in robotics. Tasks such as insertion with different objects or cloth hanging require precise control and effective modelling of complex dynamics. In this work, we frame this problem through the lens of a heterogeneous graph that comprises smaller sub-graphs, such as actuators and objects, accompanied by different edge types describing their interactions. This graph representation serves as a unified structure for both rigid and deformable objects tasks, and can be extended further to tasks comprising multiple actuators. To evaluate this setup, we present a novel and challenging reinforcement learning benchmark, including rigid insertion of diverse objects, as well as rope and cloth manipulation with multiple end-effectors. These tasks present a large search space, as both the initial and target configurations are uniformly sampled in 3D space. To address this issue, we propose a novel graph-based policy model, dubbed Heterogeneous Equivariant Policy (HEPi), utilizing $SE(3)$ equivariant message passing networks as the main backbone to exploit the geometric symmetry. In addition, by modeling explicit heterogeneity, HEPi can outperform Transformer-based and non-heterogeneous equivariant policies in terms of average returns, sample efficiency, and generalization to unseen objects.

Maximum entropy reinforcement learning (MaxEnt-RL) has become the standard approach to RL due to its beneficial exploration properties. Traditionally, policies are parameterized using Gaussian distributions, which significantly limits their representational capacity. Diffusion-based policies offer a more expressive alternative, yet integrating them into MaxEnt-RL poses challenges--primarily due to the intractability of computing their marginal entropy. To overcome this, we propose Diffusion-Based Maximum Entropy RL (DIME). DIME leverages recent advances in approximate inference with diffusion models to derive a lower bound on the maximum entropy objective. Additionally, we propose a policy iteration scheme that provably converges to the optimal diffusion policy. Our method enables the use of expressive diffusion-based policies while retaining the principled exploration benefits of MaxEnt-RL, significantly outperforming other diffusion-based methods on challenging high-dimensional control benchmarks. It is also competitive with state-of-the-art non-diffusion based RL methods while requiring fewer algorithmic design choices and smaller update-to-data ratios, reducing computational complexity.

This work introduces B-spline Movement Primitives (BMPs), a new Movement Primitive (MP) variant that leverages B-splines for motion representation. B-splines are a well-known concept in motion planning due to their ability to generate complex, smooth trajectories with only a few control points while satisfying boundary conditions, i.e., passing through a specified desired position with desired velocity. However, current usages of B-splines tend to ignore the higher-order statistics in trajectory distributions, which limits their usage in imitation learning (IL) and reinforcement learning (RL), where modeling trajectory distribution is essential. In contrast, MPs are commonly used in IL and RL for their capacity to capture trajectory likelihoods and correlations. However, MPs are constrained by their abilities to satisfy boundary conditions and usually need extra terms in learning objectives to satisfy velocity constraints. By reformulating B-splines as MPs, represented through basis functions and weight parameters, BMPs combine the strengths of both approaches, allowing B-splines to capture higher-order statistics while retaining their ability to satisfy boundary conditions. Empirical results in IL and RL demonstrate that BMPs broaden the applicability of B-splines in robot learning and offer greater expressiveness compared to existing MP variants.

An effective approach for sampling from unnormalized densities is based on the idea of gradually transporting samples from an easy prior to the complicated target distribution. Two popular methods are (1) Sequential Monte Carlo (SMC), where the transport is performed through successive annealed densities via prescribed Markov chains and resampling steps, and (2) recently developed diffusion-based sampling methods, where a learned dynamical transport is used. Despite the common goal, both approaches have different, often complementary, advantages and drawbacks. The resampling steps in SMC allow focusing on promising regions of the space, often leading to robust performance. While the algorithm enjoys asymptotic guarantees, the lack of flexible, learnable transitions can lead to slow convergence. On the other hand, diffusion-based samplers are learned and can potentially better adapt themselves to the target at hand, yet often suffer from training instabilities. In this work, we present a principled framework for combining SMC with diffusion-based samplers by viewing both methods in continuous time and considering measures on path space. This culminates in the new Sequential Controlled Langevin Diffusion (SCLD) sampling method, which is able to utilize the benefits of both methods and reaches improved performance on multiple benchmark problems, in many cases using only 10% of the training budget of previous diffusion-based samplers.