Careful robot manipulation in every-day cluttered environments requires an accurate understanding of the 3D scene, in order to grasp and place objects stably and reliably and to avoid colliding with other objects. In general, we must construct such a 3D interpretation of a complex scene based on limited input, such as a single RGB-D image. We describe SceneComplete, a system for constructing a complete, segmented, 3D model of a scene from a single view. SceneComplete is a novel pipeline for composing general-purpose pretrained perception modules (vision-language, segmentation, image-inpainting, image-to-3D, visual-descriptors and pose-estimation) to obtain highly accurate results. We demonstrate its accuracy and effectiveness with respect to ground-truth models in a large benchmark dataset and show that its accurate whole-object reconstruction enables robust grasp proposal generation, including for a dexterous hand. We release the code and additional results on our website.

The problem of mating two parts with low clearance remains difficult for autonomous robots. We present bi-level belief assembly (BILBA), a model-based planner that computes a sequence of compliant motions which can leverage contact with the environment to reduce uncertainty and perform challenging assembly tasks with low clearance. Our approach is based on first deriving candidate contact schedules from the structure of the configuration space obstacle of the parts and then finding compliant motions that achieve the desired contacts. We demonstrate that BILBA can efficiently compute robust plans on multiple simulated tasks as well as a real robot rectangular peg-in-hole insertion task.

Our aim is to learn to solve long-horizon decision-making problems in complex robotics domains given low-level skills and a handful of demonstrations containing sequences of images. To this end, we focus on learning abstract symbolic world models that facilitate zero-shot generalization to novel goals via planning. A critical component of such models is the set of symbolic predicates that define properties of and relationships between objects. In this work, we leverage pretrained vision-language models (VLMs) to propose a large set of visual predicates potentially relevant for decision-making, and to evaluate those predicates directly from camera images. At training time, we pass the proposed predicates and demonstrations into an optimization-based model-learning algorithm to obtain an abstract symbolic world model that is defined in terms of a compact subset of the proposed predicates. At test time, given a novel goal in a novel setting, we use the VLM to construct a symbolic description of the current world state, and then use a search-based planning algorithm to find a sequence of low-level skills that achieves the goal. We demonstrate empirically across experiments in both simulation and the real world that our method can generalize aggressively, applying its learned world model to solve problems with varying visual backgrounds, types, numbers, and arrangements of objects, as well as novel goals and much longer horizons than those seen at training time.

Foundation models like Vision-Language Models (VLMs) excel at commonsense vision and language tasks such as visual question answering. However, they cannot yet directly solve complex, long-horizon robot manipulation problems requiring precise continuous reasoning. Task and Motion Planning (TAMP) systems can handle long-horizon reasoning through discrete-continuous hybrid search over parameterized skills, but rely on detailed environment models and cannot interpret novel human objectives, such as arbitrary natural language goals. We propose integrating VLMs into TAMP systems by having them generate continuous (and discrete) language-parameterized constraints that enable open-world reasoning. Specifically, we use VLMs to generate continuous action constraints in the form of code, which augment traditional TAMP manipulation constraints. We also generate discrete plan constraints that parameterize actions and partially order them with respect to the continuous constraints. Experiments show that our approach, OWL-TAMP, outperforms ablations relying solely on TAMP or VLMs, as well as several VLM planning baselines from the literature, across several long-horizon manipulation tasks specified directly in natural language. We additionally demonstrate that OWL-TAMP can be deployed with an off-the-shelf TAMP system to solve challenging manipulation tasks on real-world hardware.

We present TiPToP, a modular manipulation system that integrates pretrained foundation models with a GPU-accelerated Task and Motion Planner to solve tasks directly from RGB images and natural language. TiPToP composes perception, planning, and execution modules and requires no robot training data. It can be deployed on a standard DROID setup in under an hour and adapted to new embodiments with minimal effort. We evaluate TiPToP against π_0.5-DROID, a state-of-the-art VLA fine-tuned on 350 hours of demonstrations, across two real-world DROID setups (one operated by an external team) and simulation, where TiPToP attains a higher average success rate and faster average completion time. We also evaluate on the MolmoSpaces benchmark, where TiPToP ranks first overall on pick and pick-and-place tasks among methods not trained on in-distribution data. We further show that TiPToP's modularity enables us to trace failures to specific components, revealing where to target improvements. We release TiPToP open-source to serve as a reproducible baseline and to enable further research on modular manipulation systems. Project website and code: https://tiptop-robot.github.io

Generalization to novel object configurations and instances across diverse tasks and environments is a critical challenge in robotics. Keypoint-based representations have been proven effective as a succinct representation for capturing essential object features, and for establishing a reference frame in action prediction, enabling data-efficient learning of robot skills. However, their manual design nature and reliance on additional human labels limit their scalability. In this paper, we propose KALM, a framework that leverages large pre-trained vision-language models (LMs) to automatically generate taskrelevant and cross-instance consistent keypoints. KALM distills robust and consistent keypoints across views and objects by generating proposals using LMs and verifies them against a small set of robot demonstration data. Based on the generated keypoints, we can train keypoint-conditioned policy models that predict actions in keypoint-centric frames, enabling robots to generalize effectively across varying object poses, camera views, and object instances with similar functional shapes. Our method demonstrates strong performance in the real world, adapting to different tasks and environments from only a handful of demonstrations while requiring no additional labels. Videos can be found at https://kalm-il.github.io/.

When using sampling-based motion planners such as PRMs, it is difficult to determine how many samples are required for the PRM to find a solution consistently. This is particularly relevant in Task and Motion Planning (TAMP), where many motion planning problems must be solved sequentially. We attempt to address this problem by proving an upper bound on the number of samples that are sufficient, with high probability, for a radius PRM to find a feasible solution, drawing on prior work in deterministic sampling and sample complexity theory. We also introduce a numerical algorithm that refines the bound based on the proofs of the sample complexity results we leverage. Our experiments show that our numerical algorithm is tight up to two to three orders of magnitude on planar problems for radius PRMs but becomes looser as the problem’s dimensionality increases. The numerical algorithm is empirically more useful as a heuristic for estimating the number of samples needed for a KNN PRM in low dimensions. When deployed to schedule samples for a KNN PRM in a TAMP planner, we also observe improvements in planning time in planar problems. While our experiments show that much work remains to tighten our bounds, the ideas presented in this paper are a step towards a practical sample bound.

Functional object arrangement (FORM) is the task of arranging objects to fulfill a function, for example, "set up a dining table for two." One key challenge here is that the instructions for FORM are often under-specified and do not explicitly specify the desired object goal poses. This paper presents SetItUp, a neuro-symbolic framework that learns to specify the goal poses of objects from a few training examples and a structured natural-language task specification. SetItUp uses a grounding graph, which is composed of abstract spatial relations among objects (e.g., left-of), as its intermediate representation. This decomposes the FORM problem into two stages: (i) predicting this graph among objects and (ii) predicting object poses given the grounding graph. For (i), SetItUp leverages large language models (LLMs) to induce Python programs from a task specification and a few training examples. This program can be executed to generate grounding graphs in novel scenarios. For (ii), SetItUp pre-trains a collection of diffusion models to capture primitive spatial relations and online composes these models to predict object poses based on the grounding graph. We evaluated SetItUp on a dataset spanning three distinct task families: arranging tableware on a dining table, organizing items on a bookshelf, and laying out furniture in a bedroom. Experiments show that SetItUp outperforms existing models in generating functional, physically feasible, and aesthetically pleasing object arrangements.

Recent advances in diffusion/flow-matching policies have enabled imitation learning of complex, multi-modal action trajectories. However, they are computationally expensive because they sample a trajectory of trajectories-a diffusion/flow trajectory of action trajectories. They discard intermediate action trajectories, and must wait for the sampling process to complete before any actions can be executed on the robot. We simplify diffusion/flow policies by treating action trajectories as flow trajectories. Instead of starting from pure noise, our algorithm samples from a narrow Gaussian around the last action. Then, it incrementally integrates a velocity field learned via flow matching to produce a sequence of actions that constitute a single trajectory. This enables actions to be streamed to the robot on-the-fly during the flow sampling process, and is well-suited for receding horizon policy execution. Despite streaming, our method retains the ability to model multi-modal behavior. We train flows that stabilize around demonstration trajectories to reduce distribution shift and improve imitation learning performance. Streaming flow policy outperforms prior methods while enabling faster policy execution and tighter sensorimotor loops for learning-based robot control.

Efficient exploration is critical for learning relational models in large-scale environments with complex, long-horizon tasks. Random exploration methods often collect redundant or irrelevant data, limiting their ability to learn accurate relational models of the environment. Goal-literal babbling (GLIB) improves upon random exploration by setting and planning to novel goals, but its reliance on random actions and random novel goal selection limits its scalability to larger domains. In this work, we identify the principles underlying efficient exploration in relational domains: (1) operator initialization with demonstrations that cover the distinct lifted effects necessary for planning and (2) refining preconditions to collect maximally informative transitions by selecting informative goal-action pairs and executing plans to them. To demonstrate these principles, we introduce Baking-Large, a challenging domain with extensive state-action spaces and long-horizon tasks. We evaluate methods using oracle-driven demonstrations for operator initialization and precondition-targeting guidance to efficiently gather critical transitions. Experiments show that both the oracle demonstrations and precondition-targeting oracle guidance significantly improve sample efficiency and generalization, paving the way for future methods to use these principles to efficiently learn accurate relational models in complex domains.

Partially Observable Markov Decision Processes (POMDPs) model decision making under uncertainty. While there are many approaches to approximately solving POMDPs, we aim to address the problem of learning such models. In particular, we are interested in a subclass of POMDPs wherein the components of the model, including the observation function, reward function, transition function, and initial state distribution function, can be modeled as low-complexity probabilistic graphical models in the form of a short probabilistic program. Our strategy to learn these programs uses an LLM as a prior, generating candidate probabilistic programs that are then tested against the empirical distribution and adjusted through feedback. We experiment on a number of classical toy POMDP problems, simulated MiniGrid domains, and two real mobile-base robotics search domains involving partial observability. Our results show that using an LLM to guide in the construction of a low-complexity POMDP model can be more effective than tabular POMDP learning, behavior cloning, or direct LLM planning.

Vision-Language Models (VLM) can generate plausible high-level plans when prompted with a goal, the context, an image of the scene, and any planning constraints. However, there is no guarantee that the predicted actions are geometrically and kinematically feasible for a particular robot embodiment. As a result, many prerequisite steps such as opening drawers to access objects are often omitted in their plans. Robot task and motion planners can generate motion trajectories that respect the geometric feasibility of actions and insert physically necessary actions, but do not scale to everyday problems that require common-sense knowledge and involve large state spaces comprised of many variables. We propose VLM-TAMP, a hierarchical planning algorithm that leverages a VLM to generate both semantically-meaningful and horizon-reducing intermediate subgoals that guide a task and motion planner. When a subgoal or action cannot be refined, the VLM is queried again for replanning. We evaluate VLMTAMP on kitchen tasks where a robot must accomplish cooking goals that require performing 30-50 actions in sequence and interacting with up to 21 objects. VLM-TAMP substantially outperforms baselines that rigidly and independently execute VLM-generated action sequences, both in terms of success rates (50 to 100 % versus 0 %) and average task completion percentage (72 to 100 % versus 15 to 45 %). See project site https://zt-yang.github.io/vlm-tamp-robot/ for more information.

We present a novel approach, MAGIC (manipulation analogies for generalizable intelligent contacts), for one-shot learning of manipulation strategies with fast and extensive generalization to novel objects. By leveraging a reference action trajectory, MAGIC effectively identifies similar contact points and sequences of actions on novel objects to replicate a demonstrated strategy, such as using different hooks to retrieve distant objects of different shapes and sizes. Our method is based on a two-stage contact-point matching process that combines global shape matching using pretrained neural features with local curvature analysis to ensure precise and physically plausible contact points. We experiment with three tasks including scooping, hanging, and hooking objects. MAGIC demonstrates superior performance over existing methods, achieving significant improvements in runtime speed and generalization to different object categories. Website: https://magic-2024.github.io/ .

Humans can learn a new manipulation task from one or two demonstrations and then perform it in a new room, with new objects, under new constraints. Modern robot imitation learning, in contrast, typically needs hundreds to thousands of demonstrations and still degrades under modest shifts in layout, geometry, object set or task constraints. We argue this gap is not just about data, but also about the level of abstraction at which learning occurs; generalization requires inferring the latent intent underlying why a demonstrator behaved in a certain way, rather than reproducing how they moved. We present Rational Inverse Reasoning (RIR), which casts few-shot imitation as inference over latent explanation programs: compact, executable descriptions of intent that map an object-centric scene to a structured task-and-motion-planning (TAMP) specification of goals, subgoals and constraints. A vision-language model proposes candidate programs, and a hierarchical planner supplies a bounded-rational likelihood. By combining VLM program proposals, and planner-grounded feedback, RIR iteratively refines the candidate set to approximate a posterior over concise, executable programs. On a 2D reasoning benchmark and a real Franka FR3, RIR recovers transferable task structure from as little as one demonstration. Generalizing to substantially new layouts and object sets, RIR outperforms VLM-planning baselines that lack explicit rationality and planning-grounded inference, increasing downstream success rate by 34 and 28 percentage points in the one- and three-shot settings.

We introduce uncertainty-aware object instance segmentation (UncOS) and demonstrate its usefulness for embodied interactive segmentation. To deal with uncertainty in robot perception, we propose a method for generating a hypothesis distribution of object segmentation. We obtain a set of region-factored segmentation hypotheses together with confidence estimates by making multiple queries of large pre-trained models. This process can produce segmentation results that achieve state-of-the-art performance on unseen object segmentation problems. The output can also serve as input to a belief-driven process for selecting robot actions to perturb the scene to reduce ambiguity. We demonstrate the effectiveness of this method in real-robot experiments. Website: https://sites.google.com/view/embodied-uncertain-seg

Multi-step forceful manipulation tasks, such as opening a push-and-twist childproof bottle, require a robot to make various planning choices that are substantially impacted by the requirement to exert force during the task. The robot must reason over discrete and continuous choices relating to the sequence of actions, such as whether to pick up an object, and the parameters of each of those actions, such how to grasp the object. To enable planning and executing forceful manipulation, we augment an existing task and motion planner with constraints that explicitly consider torque and frictional limits, captured through the proposed forceful kinematic chain constraint. In three domains, opening a childproof bottle, twisting a nut and cutting a vegetable, we demonstrate how the system selects from among a combinatorial set of strategies.We also show how cost-sensitive planning can be used to find strategies and parameters that are robust to uncertainty in the physical parameters.

One promising approach towards effective robot decision making in complex, long-horizon tasks is to sequence together parameterized skills. We consider a setting where a robot is initially equipped with (1) a library of parameterized skills, (2) an AI planner for sequencing together the skills given a goal, and (3) a very general prior distribution for selecting skill parameters. Once deployed, the robot should rapidly and autonomously learn to improve its performance by specializing its skill parameter selection policy to the particular objects, goals, and constraints in its environment. In this work, we focus on the active learning problem of choosing which skills to practice to maximize expected future task success. We propose that the robot should estimate the competence of each skill, extrapolate the competence (asking: "how much would the competence improve through practice?"), and situate the skill in the task distribution through competence-aware planning. This approach is implemented within a fully autonomous system where the robot repeatedly plans, practices, and learns without any environment resets. Through experiments in simulation, we find that our approach learns effective parameter policies more sample-efficiently than several baselines. Experiments in the real-world demonstrate our approach's ability to handle noise from perception and control and improve the robot's ability to solve two long-horizon mobile-manipulation tasks after a few hours of autonomous practice.

Manipulation of large objects over long horizons (such as carts in a warehouse) is an essential skill for deployable robotic systems. Large objects require mobile manipulation which involves simultaneous manipulation, navigation, and movement with the object in tow. In many real-world situations, object dynamics are incredibly complex, such as the interaction of an office chair (with a rotating base and five caster wheels) and the ground. We present a hierarchical algorithm for long-horizon robot manipulation problems in which the dynamics are partially unknown. We observe that diffusion-based behavior cloning is highly effective for short-horizon problems with unknown dynamics, so we decompose the problem into an abstract high-level, obstacle-aware motion-planning problem that produces a waypoint sequence. We use a short-horizon, relative-motion diffusion policy to achieve the waypoints in sequence. We train mobile manipulation policies on a Spot robot that has to push and pull an office chair. Our hierarchical manipulation policy performs consistently better, especially when the horizon increases, compared to a diffusion policy trained on long-horizon demonstrations or motion planning assuming a rigidly-attached object (success rate of 8 (versus 0 and 5 respectively) out of 10 runs). Importantly, our learned policy generalizes to new layouts, grasps, chairs, and flooring that induces more friction, without any further training, showing promise for other complex mobile manipulation problems. Project Page: https://yravan.github.io/plannerorderedpolicy/

This paper studies the challenge of developing robots capable of understanding under-specified instructions for creating functional object arrangements, such as "set up a dining table for two"; previous arrangement approaches have focused on much more explicit instructions, such as "put object A on the table." We introduce a framework, SetItUp, for learning to interpret under-specified instructions. SetItUp takes a small number of training examples and a human-crafted program sketch to uncover arrangement rules for specific scene types. By leveraging an intermediate graph-like representation of abstract spatial relationships among objects, SetItUp decomposes the arrangement problem into two subproblems: i) learning the arrangement patterns from limited data and ii) grounding these abstract relationships into object poses. SetItUp leverages large language models (LLMs) to propose the abstract spatial relationships among objects in novel scenes as the constraints to be satisfied; then, it composes a library of diffusion models associated with these abstract relationships to find object poses that satisfy the constraints. We validate our framework on a dataset comprising study desks, dining tables, and coffee tables, with the results showing superior performance in generating physically plausible, functional, and aesthetically pleasing object arrangements compared to existing models.

The real world is unpredictable. Therefore, to solve long-horizon decision-making problems with autonomous robots, we must construct agents that are capable of adapting to changes in the environment during deployment. Model-based planning approaches can enable robots to solve complex, long-horizon tasks in a variety of environments. However, such approaches tend to be brittle when deployed into an environment featuring a novel situation that their underlying model does not account for. In this work, we propose to learn a “bridge policy” via Reinforcement Learning (RL) to adapt to such novelties. We introduce a simple formulation for such learning, where the RL problem is constructed with a special “CallPlanner” action that terminates the bridge policy and hands control of the agent back to the planner. This allows the RL policy to learn the set of states in which querying the planner and following the returned plan will achieve the goal. We show that this formulation enables the agent to rapidly learn by leveraging the planner's knowledge to avoid challenging long-horizon exploration caused by sparse reward. In experiments across three different simulated domains of varying complexity, we demonstrate that our approach is able to learn policies that adapt to novelty more efficiently than several baselines, including a pure RL baseline. We also demonstrate that the learned bridge policy is generalizable in that it can be combined with the planner to enable the agent to solve more complex tasks with multiple instances of the encountered novelty.