Whole-Body Mobile Manipulation using Offline Reinforcement Learning on Sub-optimal Controllers

Mobile Manipulation (MoMa) of articulated objects, such as opening doors, drawers, and cupboards, demands simultaneous, whole-body coordination between a robot's base and arms. Classical whole-body controllers (WBCs) can solve such problems via hierarchical optimization, but require extensive hand-tuned optimization and remain brittle. Learning-based methods, on the other hand, show strong generalization capabilities but typically rely on expensive whole-body teleoperation data or heavy reward engineering. We observe that even a sub-optimal WBC is a powerful structural prior: it can be used to collect data in a constrained, task-relevant region of the state-action space, and its behavior can still be improved upon using offline reinforcement learning. Building on this, we propose WHOLE-MoMa, a two-stage pipeline that first generates diverse demonstrations by randomizing a lightweight WBC, and then applies offline RL to identify and stitch together improved behaviors via a reward signal. To support the expressive action-chunked diffusion policies needed for complex coordination tasks, we extend offline implicit Q-learning with Q-chunking for chunk-level critic evaluation and advantage-weighted policy extraction. On three tasks of increasing difficulty using a TIAGo++ mobile manipulator in simulation, WHOLE-MoMa significantly outperforms WBC, behavior cloning, and several offline RL baselines. Policies transfer directly to the real robot without finetuning, achieving 80% success in bimanual drawer manipulation and 68% in simultaneous cupboard opening and object placement, all without any teleoperated or real-world training data.

Task-Aware Bimanual Affordance Prediction via VLM-Guided Semantic-Geometric Reasoning

Bimanual manipulation requires reasoning about where to interact with an object and which arm should perform each action, a joint affordance localization and arm allocation problem that geometry-only planners cannot resolve without semantic understanding of task intent. Existing approaches either treat affordance prediction as coarse part segmentation or rely on geometric heuristics for arm assignment, failing to jointly reason about task-relevant contact regions and arm allocation. We reframe bimanual manipulation as a joint affordance localization and arm allocation problem and propose a hierarchical framework for task-aware bimanual affordance prediction that leverages a Vision-Language Model (VLM) to generalize across object categories and task descriptions without requiring category-specific training. Our approach fuses multi-view RGB-D observations into a consistent 3D scene representation and generates global 6-DoF grasp candidates, which are then spatially and semantically filtered by querying the VLM for task-relevant affordance regions on each object, as well as for arm allocation to the individual objects, thereby ensuring geometric validity while respecting task semantics. We evaluate our method on a dual-arm platform across nine real-world manipulation tasks spanning four categories: parallel manipulation, coordinated stabilization, tool use, and human handover. Our approach achieves consistently higher task success rates than geometric and semantic baselines for task-oriented grasping, demonstrating that explicit semantic reasoning over affordances and arm allocation helps enable reliable bimanual manipulation in unstructured environments.

MolmoB0T: Large-Scale Simulation Enables Zero-Shot Manipulation

A prevailing view in robot learning is that simulation alone is not enough; effective sim-to-real transfer is widely believed to require at least some real-world data collection or task-specific fine-tuning to bridge the gap between simulated and physical environments. We challenge that assumption. With sufficiently large-scale and diverse simulated synthetic training data, we show that zero-shot transfer to the real world is not only possible, but effective for both static and mobile manipulation. We introduce MolmoBot-Engine, a fully open-source pipeline for procedural data generation across robots, tasks, and diverse simulated environments in MolmoSpaces. With it, we release MolmoBot-Data, a dataset of 1.8 million expert trajectories for articulated object manipulation and pick-and-place tasks. We train three policy classes: MolmoBot, a Molmo2-based multi-frame vision-language model with a flow-matching action head; MolmoBot-Pi0, which replicates the $π_0$ architecture to enable direct comparison; and MolmoBot-SPOC, a lightweight policy suitable for edge deployment and amenable to RL fine-tuning. We evaluate on two robotic platforms: the Franka FR3 for tabletop manipulation tasks and the Rainbow Robotics RB-Y1 mobile manipulator for door opening, drawer manipulation, cabinet interaction, and mobile pick-and-place. Without any real-world fine-tuning, our policies achieve zero-shot transfer to unseen objects and environments. On tabletop pick-and-place, MolmoBot achieves a success rate of 79.2% in real world evaluations across 4 settings, outperforming $π_{0.5}$ at 39.2%. Our results demonstrate that procedural environment generation combined with diverse articulated assets can produce robust manipulation policies that generalize broadly to the real world. Technical Blog: https://allenai.org/blog/molmobot-robot-manipulation

Stein Variational Ergodic Surface Coverage with SE(3) Constraints

Surface manipulation tasks require robots to generate trajectories that comprehensively cover complex 3D surfaces while maintaining precise end-effector poses. Existing ergodic trajectory optimization (TO) methods demonstrate success in coverage tasks, while struggling with point-cloud targets due to the nonconvex optimization landscapes and the inadequate handling of SE(3) constraints in sampling-as-optimization (SAO) techniques. In this work, we introduce a preconditioned SE(3) Stein Variational Gradient Descent (SVGD) approach for SAO ergodic trajectory generation. Our proposed approach comprises multiple innovations. First, we reformulate point-cloud ergodic coverage as a manifold-aware sampling problem. Second, we derive SE(3)-specific SVGD particle updates, and, third, we develop a preconditioner to accelerate TO convergence. Our sampling-based framework consistently identifies superior local optima compared to strong optimization-based and SAO baselines while preserving the SE(3) geometric structure. Experiments on a 3D point-cloud surface coverage benchmark and robotic surface drawing tasks demonstrate that our method achieves superior coverage quality with tractable computation in our setting relative to existing TO and SAO approaches, and is validated in real-world robot experiments.

Robot-DIFT: Distilling Diffusion Features for Geometrically Consistent Visuomotor Control

We hypothesize that a key bottleneck in generalizable robot manipulation is not solely data scale or policy capacity, but a structural mismatch between current visual backbones and the physical requirements of closed-loop control. While state-of-the-art vision encoders (including those used in VLAs) optimize for semantic invariance to stabilize classification, manipulation typically demands geometric sensitivity the ability to map millimeter-level pose shifts to predictable feature changes. Their discriminative objective creates a "blind spot" for fine-grained control, whereas generative diffusion models inherently encode geometric dependencies within their latent manifolds, encouraging the preservation of dense multi-scale spatial structure. However, directly deploying stochastic diffusion features for control is hindered by stochastic instability, inference latency, and representation drift during fine-tuning. To bridge this gap, we propose Robot-DIFT, a framework that decouples the source of geometric information from the process of inference via Manifold Distillation. By distilling a frozen diffusion teacher into a deterministic Spatial-Semantic Feature Pyramid Network (S2-FPN), we retain the rich geometric priors of the generative model while ensuring temporal stability, real-time execution, and robustness against drift. Pretrained on the large-scale DROID dataset, Robot-DIFT demonstrates superior geometric consistency and control performance compared to leading discriminative baselines, supporting the view that how a model learns to see dictates how well it can learn to act.

MolmoSpaces: A Large-Scale Open Ecosystem for Robot Navigation and Manipulation

Deploying robots at scale demands robustness to the long tail of everyday situations. The countless variations in scene layout, object geometry, and task specifications that characterize real environments are vast and underrepresented in existing robot benchmarks. Measuring this level of generalization requires infrastructure at a scale and diversity that physical evaluation alone cannot provide. We introduce MolmoSpaces, a fully open ecosystem to support large-scale benchmarking of robot policies. MolmoSpaces consists of over 230k diverse indoor environments, ranging from handcrafted household scenes to procedurally generated multiroom houses, populated with 130k richly annotated object assets, including 48k manipulable objects with 42M stable grasps. Crucially, these environments are simulator-agnostic, supporting popular options such as MuJoCo, Isaac, and ManiSkill. The ecosystem supports the full spectrum of embodied tasks: static and mobile manipulation, navigation, and multiroom long-horizon tasks requiring coordinated perception, planning, and interaction across entire indoor environments. We also design MolmoSpaces-Bench, a benchmark suite of 8 tasks in which robots interact with our diverse scenes and richly annotated objects. Our experiments show MolmoSpaces-Bench exhibits strong sim-to-real correlation (R = 0.96, \r{ho} = 0.98), confirm newer and stronger zero-shot policies outperform earlier versions in our benchmarks, and identify key sensitivities to prompt phrasing, initial joint positions, and camera occlusion. Through MolmoSpaces and its open-source assets and tooling, we provide a foundation for scalable data generation, policy training, and benchmark creation for robot learning research.

Learning Semantic-Geometric Task Graph-Representations from Human Demonstrations

Learning structured task representations from human demonstrations is essential for understanding long-horizon manipulation behaviors, particularly in bimanual settings where action ordering, object involvement, and interaction geometry can vary significantly. A key challenge lies in jointly capturing the discrete semantic structure of tasks and the temporal evolution of object-centric geometric relations in a form that supports reasoning over task progression. In this work, we introduce a semantic-geometric task graph-representation that encodes object identities, inter-object relations, and their temporal geometric evolution from human demonstrations. Building on this formulation, we propose a learning framework that combines a Message Passing Neural Network (MPNN) encoder with a Transformer-based decoder, decoupling scene representation learning from action-conditioned reasoning about task progression. The encoder operates solely on temporal scene graphs to learn structured representations, while the decoder conditions on action-context to predict future action sequences, associated objects, and object motions over extended time horizons. Through extensive evaluation on human demonstration datasets, we show that semantic-geometric task graph-representations are particularly beneficial for tasks with high action and object variability, where simpler sequence-based models struggle to capture task progression. Finally, we demonstrate that task graph representations can be transferred to a physical bimanual robot and used for online action selection, highlighting their potential as reusable task abstractions for downstream decision-making in manipulation systems.

Self-Supervised Multisensory Pretraining for Contact-Rich Robot Reinforcement Learning

Effective contact-rich manipulation requires robots to synergistically leverage vision, force, and proprioception. However, Reinforcement Learning agents struggle to learn in such multisensory settings, especially amidst sensory noise and dynamic changes. We propose MultiSensory Dynamic Pretraining (MSDP), a novel framework for learning expressive multisensory representations tailored for task-oriented policy learning. MSDP is based on masked autoencoding and trains a transformer-based encoder by reconstructing multisensory observations from only a subset of sensor embeddings, leading to cross-modal prediction and sensor fusion. For downstream policy learning, we introduce a novel asymmetric architecture, where a cross-attention mechanism allows the critic to extract dynamic, task-specific features from the frozen embeddings, while the actor receives a stable pooled representation to guide its actions. Our method demonstrates accelerated learning and robust performance under diverse perturbations, including sensor noise, and changes in object dynamics. Evaluations in multiple challenging, contact-rich robot manipulation tasks in simulation and the real world showcase the effectiveness of MSDP. Our approach exhibits strong robustness to perturbations and achieves high success rates on the real robot with as few as 6,000 online interactions, offering a simple yet powerful solution for complex multisensory robotic control.

Adaptive Diffusion Constrained Sampling for Bimanual Robot Manipulation

Coordinated multi-arm manipulation requires satisfying multiple simultaneous geometric constraints across high-dimensional configuration spaces, which poses a significant challenge for traditional planning and control methods. In this work, we propose Adaptive Diffusion Constrained Sampling (ADCS), a generative framework that flexibly integrates both equality (e.g., relative and absolute pose constraints) and structured inequality constraints (e.g., proximity to object surfaces) into an energy-based diffusion model. Equality constraints are modeled using dedicated energy networks trained on pose differences in Lie algebra space, while inequality constraints are represented via Signed Distance Functions (SDFs) and encoded into learned constraint embeddings, allowing the model to reason about complex spatial regions. A key innovation of our method is a Transformer-based architecture that learns to weight constraint-specific energy functions at inference time, enabling flexible and context-aware constraint integration. Moreover, we adopt a two-phase sampling strategy that improves precision and sample diversity by combining Langevin dynamics with resampling and density-aware re-weighting. Experimental results on dual-arm manipulation tasks show that ADCS significantly improves sample diversity and generalization across settings demanding precise coordination and adaptive constraint handling.

Morphologically Symmetric Reinforcement Learning for Ambidextrous Bimanual Manipulation

Humans naturally exhibit bilateral symmetry in their gross manipulation skills, effortlessly mirroring simple actions between left and right hands. Bimanual robots-which also feature bilateral symmetry-should similarly exploit this property to perform tasks with either hand. Unlike humans, who often favor a dominant hand for fine dexterous skills, robots should ideally execute ambidextrous manipulation with equal proficiency. To this end, we introduce SYMDEX (SYMmetric DEXterity), a reinforcement learning framework for ambidextrous bi-manipulation that leverages the robot's inherent bilateral symmetry as an inductive bias. SYMDEX decomposes complex bimanual manipulation tasks into per-hand subtasks and trains dedicated policies for each. By exploiting bilateral symmetry via equivariant neural networks, experience from one arm is inherently leveraged by the opposite arm. We then distill the subtask policies into a global ambidextrous policy that is independent of the hand-task assignment. We evaluate SYMDEX on six challenging simulated manipulation tasks and demonstrate successful real-world deployment on two of them. Our approach strongly outperforms baselines on complex task in which the left and right hands perform different roles. We further demonstrate SYMDEX's scalability by extending it to a four-arm manipulation setup, where our symmetry-aware policies enable effective multi-arm collaboration and coordination. Our results highlight how structural symmetry as inductive bias in policy learning enhances sample efficiency, robustness, and generalization across diverse dexterous manipulation tasks.