SEARCH · Search arXiv

Results for “robot learning”

Search indexed arXiv papers on artificial intelligence, large language models, computer vision and robotics. Read source abstracts and follow links to arXiv.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,531 records · Page 85Linked to original sources

arXiv2026

Learning to Plan in Human-Robot Collaboration: Multimodal Reinforcement Learning for Adaptive Interaction

Robot assistants for older adults and people with disabilities need to perform collaborative tasks with users effectively. The core component of these systems is an interaction manager whose job is to observe and assess the task and infer the state of the human and their intent for the robot to choose the best course of action. Due to the sparseness of the data in this domain, the policy for such multimodal systems is often crafted by hand; as the complexity of interactions grows, this process is not scalable. This paper proposes a reinforcement learning (RL) approach to automatically generate the multimodal policy of the robot. Our system focuses on a realistic scenario where a robot assists a user in locating objects within a home environment, managing multimodal signals, including language and physical actions, to select the best action. In contrast to traditional dialog systems, our agent is trained with a simulator that uses human data and can deal with multiple modalities. We use a simple high-level reward function that needs no fine-tuning and enforce some preconditions to speed up the training process. A human study evaluating the system in a real-world setting demonstrates promising results, indicating high usability and effective task completion. This RL-based approach offers a scalable and interpretable alternative for designing interaction managers in multimodal human-robot collaborations.

cs.RO↗

arXiv2025

Entropy-Controlled Intrinsic Motivation Reinforcement Learning for Quadruped Robot Locomotion in Complex Terrains

Learning is the basis of both biological and artificial systems when it comes to mimicking intelligent behaviors. From the classical PPO (Proximal Policy Optimization), there is a series of deep reinforcement learning algorithms which are widely used in training locomotion policies for quadrupedal robots because of their stability and sample efficiency. However, among all these variants, experiments and simulations often converge prematurely, leading to suboptimal locomotion and reduced task performance. Therefore, in this paper, we introduce Entropy-Controlled Intrinsic Motivation (ECIM), an entropy-based reinforcement learning algorithm in contrast with the PPO series, that can reduce premature convergence by combining intrinsic motivation with adaptive exploration. For experiments, in order to parallel with other baselines, we chose to apply it in Isaac Gym across six terrain categories: upward slopes, downward slopes, uneven rough terrain, ascending stairs, descending stairs, and flat ground as widely used. For comparison, our experiments consistently achieve better performance: task rewards increase by 4--12%, peak body pitch oscillation is reduced by 23--29%, joint acceleration decreases by 20--32%, and joint torque consumption declines by 11--20%. Overall, our model ECIM, by combining entropy control and intrinsic motivation control, achieves better results in stability across different terrains for quadrupedal locomotion, and at the same time reduces energetic cost and makes it a practical choice for complex robotic control tasks.

cs.RO↗

arXiv2026

VGFM: Expressive Robot Policies via Dense Value Guidance in Flow Matching

Recent robot learning paradigms increasingly rely on large offline datasets of robotic interactions to train control policies. Expressive generative models enable rich and multimodal action representations, expanding the capability of this paradigm for complex robotic control. However, policy improvement with multi-step generative actors remains challenging. In offline reinforcement learning (RL), incorporating value-based objectives along generative trajectories often introduces substantial training complexity, including backpropagation through time (BPTT), auxiliary architectures, or distillation losses. We propose Value-Guided Flow Matching (VGFM), a scalable offline RL framework that enables dense value-guided shaping within a flow-based policy while avoiding BPTT and additional algorithmic overhead. VGFM parameterizes the policy as a conditional flow-matching model in action (x-prediction) space, ensuring that each intermediate flow step produces a valid robot action that can be directly evaluated by a standard offline RL critic. This design allows value guidance to be applied at randomly sampled flow times without differentiating through the entire generative trajectory, while preserving inference-time flexibility by varying the discretization of the underlying flow ODE without retraining. Evaluated on robotic locomotion and manipulation tasks in OGBench, VGFM achieves strong performance across a wide range of tasks under rigorous evaluation protocols. With minimal hyperparameter tuning, these results demonstrate that VGFM provides a simple, scalable, and effective approach for expressive policy learning in long-horizon, goal-oriented robotic control.

cs.RO↗

arXiv2021

DRL: Deep Reinforcement Learning for Intelligent Robot Control -- Concept, Literature, and Future

Combination of machine learning (for generating machine intelligence), computer vision (for better environment perception), and robotic systems (for controlled environment interaction) motivates this work toward proposing a vision-based learning framework for intelligent robot control as the ultimate goal (vision-based learning robot). This work specifically introduces deep reinforcement learning as the the learning framework, a General-purpose framework for AI (AGI) meaning application-independent and platform-independent. In terms of robot control, this framework is proposing specifically a high-level control architecture independent of the low-level control, meaning these two required level of control can be developed separately from each other. In this aspect, the high-level control creates the required intelligence for the control of the platform using the recorded low-level controlling data from that same platform generated by a trainer. The recorded low-level controlling data is simply indicating the successful and failed experiences or sequences of experiments conducted by a trainer using the same robotic platform. The sequences of the recorded data are composed of observation data (input sensor), generated reward (feedback value) and action data (output controller). For experimental platform and experiments, vision sensors are used for perception of the environment, different kinematic controllers create the required motion commands based on the platform application, deep learning approaches generate the required intelligence, and finally reinforcement learning techniques incrementally improve the generated intelligence until the mission is accomplished by the robot.

cs.RO↗

arXiv2026

Agentic Scene Policies

Designing or learning robot policies that generalize zero-shot across a range of language instructions and objects is a core problem in robotics. Vision-Language-Action models (VLAs) learn such policies end-to-end by repurposing existing Vision-Language Models (VLMs), but generalization to new instructions and objects remains challenging. An alternative is to implement a modular policy by leveraging an explicit VLM-based 3D scene representation and motion planning. While modular policies show strong zero-shot potential, they typically retrieve objects based on semantics without explicit spatial reasoning, severely restricting their overall grounding capabilities. They also interact with objects using basic grasping and navigation skills. In this work, we address these limitations by unifying grounding capabilities and robot skills in a single agentic action space through a scene-agent tool interface. By leveraging part-level affordances, our skills generalize across diverse objects and enable zero-shot interactions such as unplugging chargers and opening drawers. We name the resulting framework Agentic Scene Policies (ASP). Through extensive real-world experiments, we show how ASP consistently outperforms leading VLAs in the zero-shot setting. We also demonstrate the extensibility of our framework by introducing a mobile version of ASP to tackle room-level queries. See our project page (https://montrealrobotics.ca/agentic-scene-policies.github.io/) for more results.

cs.RO↗

arXiv2025

Spatial-Aware VLA Pretraining through Visual-Physical Alignment from Human Videos

Vision-Language-Action (VLA) models provide a promising paradigm for robot learning by integrating visual perception with language-guided policy learning. However, most existing approaches rely on 2D visual inputs to perform actions in 3D physical environments, creating a significant gap between perception and action grounding. To bridge this gap, we propose a Spatial-Aware VLA Pretraining paradigm that performs explicit alignment between visual space and physical space during pretraining, enabling models to acquire 3D spatial understanding before robot policy learning. Starting from pretrained vision-language models, we leverage large-scale human demonstration videos to extract 3D visual and 3D action annotations, forming a new source of supervision that aligns 2D visual observations with 3D spatial reasoning. We instantiate this paradigm with VIPA-VLA, a dual-encoder architecture that incorporates a 3D visual encoder to augment semantic visual representations with 3D-aware features. When adapted to downstream robot tasks, VIPA-VLA achieves significantly improved grounding between 2D vision and 3D action, resulting in more robust and generalizable robotic policies.

cs.RO↗

arXiv2026

MARS Policy: Multimodality Only When It Matters

Imitation learning has become a cornerstone for solving complex robotic manipulation tasks. In particular, multimodality, which enables robots to capture diverse yet valid behavioral patterns, has driven the rapid emergence of generative policies as a dominant paradigm in robot learning. However, achieving such multimodality typically relies on stochastic noise initialization and iterative denoising procedures, resulting in substantial training complexity and low inference efficiency. Meanwhile, not all phases of a robotic task inherently require behavioral diversity. Motivated by this insight, we propose the Modality-Adaptive Robot Sampling (MARS) policy, which adaptively invokes tailored stochasticity only when it is truly beneficial, while reverting to an efficient deterministic learning during single-modal phases. In other words, the proper amount of noise is injected only at the proper time. By selectively activating multimodal generation, MARS policy bridges the gap between the multimodal capability of generative policies and the superior training and inference efficiency of deterministic models. Empirical studies across 8 simulated and 4 real-world tasks demonstrate that MARS exhibits robust multimodal expressivity and high efficiency, with a 16.67% success rate improvement and an 83.20% inference latency reduction in real-world tests. Counterintuitively, MARS also outpaces deterministic policies in training efficiency on near-deterministic tasks by more effectively modeling nuanced action diversity.

cs.RO↗

arXiv2026

Stability-Guided Exploration for Diverse Motion Generation

Scaling up datasets is highly effective in improving the performance of deep learning models, including in the field of robot learning. However, data collection still proves to be a bottleneck. Approaches relying on collecting human demonstrations are labor-intensive and inherently limited: they tend to be narrow, task-specific, and fail to adequately explore the full space of feasible states. Synthetic data generation could remedy this, but current techniques mostly rely on local trajectory optimization and fail to find diverse solutions. In this work, we propose a novel method capable of finding diverse long-horizon manipulations through black-box simulation. We achieve this by combining an RRT-style search with sampling-based MPC, together with a novel sampling scheme that guides the exploration toward stable configurations. Specifically, we sample from a manifold of stable states while growing a search tree directly through simulation, without restricting the planner to purely stable motions. We demonstrate the method's ability to discover diverse manipulation strategies, including pushing, grasping, pivoting, throwing, and tool use, across different robot morphologies, without task-specific guidance.

cs.RO↗

arXiv2024

PREDILECT: Preferences Delineated with Zero-Shot Language-based Reasoning in Reinforcement Learning

Preference-based reinforcement learning (RL) has emerged as a new field in robot learning, where humans play a pivotal role in shaping robot behavior by expressing preferences on different sequences of state-action pairs. However, formulating realistic policies for robots demands responses from humans to an extensive array of queries. In this work, we approach the sample-efficiency challenge by expanding the information collected per query to contain both preferences and optional text prompting. To accomplish this, we leverage the zero-shot capabilities of a large language model (LLM) to reason from the text provided by humans. To accommodate the additional query information, we reformulate the reward learning objectives to contain flexible highlights -- state-action pairs that contain relatively high information and are related to the features processed in a zero-shot fashion from a pretrained LLM. In both a simulated scenario and a user study, we reveal the effectiveness of our work by analyzing the feedback and its implications. Additionally, the collective feedback collected serves to train a robot on socially compliant trajectories in a simulated social navigation landscape. We provide video examples of the trained policies at https://sites.google.com/view/rl-predilect

cs.RO↗

arXiv2025

Integrating Offline Pre-Training with Online Fine-Tuning: A Reinforcement Learning Approach for Robot Social Navigation

Offline reinforcement learning (RL) has emerged as a promising framework for addressing robot social navigation challenges. However, inherent uncertainties in pedestrian behavior and limited environmental interaction during training often lead to suboptimal exploration and distributional shifts between offline training and online deployment. To overcome these limitations, this paper proposes a novel offline-to-online fine-tuning RL algorithm for robot social navigation by integrating Return-to-Go (RTG) prediction into a causal Transformer architecture. Our algorithm features a spatiotem-poral fusion model designed to precisely estimate RTG values in real-time by jointly encoding temporal pedestrian motion patterns and spatial crowd dynamics. This RTG prediction framework mitigates distribution shift by aligning offline policy training with online environmental interactions. Furthermore, a hybrid offline-online experience sampling mechanism is built to stabilize policy updates during fine-tuning, ensuring balanced integration of pre-trained knowledge and real-time adaptation. Extensive experiments in simulated social navigation environments demonstrate that our method achieves a higher success rate and lower collision rate compared to state-of-the-art baselines. These results underscore the efficacy of our algorithm in enhancing navigation policy robustness and adaptability. This work paves the way for more reliable and adaptive robotic navigation systems in real-world applications.

cs.RO↗

arXiv2021

I am Robot: Neuromuscular Reinforcement Learning to Actuate Human Limbs through Functional Electrical Stimulation

Human movement disorders or paralysis lead to the loss of control of muscle activation and thus motor control. Functional Electrical Stimulation (FES) is an established and safe technique for contracting muscles by stimulating the skin above a muscle to induce its contraction. However, an open challenge remains on how to restore motor abilities to human limbs through FES, as the problem of controlling the stimulation is unclear. We are taking a robotics perspective on this problem, by developing robot learning algorithms that control the ultimate humanoid robot, the human body, through electrical muscle stimulation. Human muscles are not trivial to control as actuators due to their force production being non-stationary as a result of fatigue and other internal state changes, in contrast to robot actuators which are well-understood and stationary over broad operation ranges. We present our Deep Reinforcement Learning approach to the control of human muscles with FES, using a recurrent neural network for dynamic state representation, to overcome the unobserved elements of the behaviour of human muscles under external stimulation. We demonstrate our technique both in neuromuscular simulations but also experimentally on a human. Our results show that our controller can learn to manipulate human muscles, applying appropriate levels of stimulation to achieve the given tasks while compensating for advancing muscle fatigue which arises throughout the tasks. Additionally, our technique can learn quickly enough to be implemented in real-world human-in-the-loop settings.

cs.RO↗

arXiv2021

Distilled Domain Randomization

Deep reinforcement learning is an effective tool to learn robot control policies from scratch. However, these methods are notorious for the enormous amount of required training data which is prohibitively expensive to collect on real robots. A highly popular alternative is to learn from simulations, allowing to generate the data much faster, safer, and cheaper. Since all simulators are mere models of reality, there are inevitable differences between the simulated and the real data, often referenced as the 'reality gap'. To bridge this gap, many approaches learn one policy from a distribution over simulators. In this paper, we propose to combine reinforcement learning from randomized physics simulations with policy distillation. Our algorithm, called Distilled Domain Randomization (DiDoR), distills so-called teacher policies, which are experts on domains that have been sampled initially, into a student policy that is later deployed. This way, DiDoR learns controllers which transfer directly from simulation to reality, i.e., without requiring data from the target domain. We compare DiDoR against three baselines in three sim-to-sim as well as two sim-to-real experiments. Our results show that the target domain performance of policies trained with DiDoR is en par or better than the baselines'. Moreover, our approach neither increases the required memory capacity nor the time to compute an action, which may well be a point of failure for successfully deploying the learned controller.

cs.LG↗

arXiv2024

SpotLight: Robotic Scene Understanding through Interaction and Affordance Detection

Despite increasing research efforts on household robotics, robots intended for deployment in domestic settings still struggle with more complex tasks such as interacting with functional elements like drawers or light switches, largely due to limited task-specific understanding and interaction capabilities. These tasks require not only detection and pose estimation but also an understanding of the affordances these elements provide. To address these challenges and enhance robotic scene understanding, we introduce SpotLight: A comprehensive framework for robotic interaction with functional elements, specifically light switches. Furthermore, this framework enables robots to improve their environmental understanding through interaction. Leveraging VLM-based affordance prediction to estimate motion primitives for light switch interaction, we achieve up to 84% operation success in real world experiments. We further introduce a specialized dataset containing 715 images as well as a custom detection model for light switch detection. We demonstrate how the framework can facilitate robot learning through physical interaction by having the robot explore the environment and discover previously unknown relationships in a scene graph representation. Lastly, we propose an extension to the framework to accommodate other functional interactions such as swing doors, showcasing its flexibility. Videos and Code: timengelbracht.github.io/SpotLight/

cs.RO↗

arXiv2025

GRAPPA: Generalizing and Adapting Robot Policies via Online Agentic Guidance

Robot learning approaches such as behavior cloning and reinforcement learning have shown great promise in synthesizing robot skills from human demonstrations in specific environments. However, these approaches often require task-specific demonstrations or designing complex simulation environments, which limits the development of generalizable and robust policies for unseen real-world settings. Recent advances in the use of foundation models for robotics (e.g., LLMs, VLMs) have shown great potential in enabling systems to understand the semantics in the world from large-scale internet data. However, it remains an open challenge to use this knowledge to enable robotic systems to understand the underlying dynamics of the world, to generalize policies across different tasks, and to adapt policies to new environments. To alleviate these limitations, we propose an agentic framework for robot self-guidance and self-improvement, which consists of a set of role-specialized conversational agents, such as a high-level advisor, a grounding agent, a monitoring agent, and a robotic agent. Our framework iteratively grounds a base robot policy to relevant objects in the environment and uses visuomotor cues to shift the action distribution of the policy to more desirable states, online, while remaining agnostic to the subjective configuration of a given robot hardware platform. We demonstrate that our approach can effectively guide manipulation policies to achieve significantly higher success rates, both in simulation and in real-world experiments, without the need for additional human demonstrations or extensive exploration. Code and videos available at: https://agenticrobots.github.io

cs.RO↗

arXiv2026

E0.5: From Reconstructive to Future-Predictive 3D Foundations for Robotic World-Action Modeling

World-action models (WAMs) offer a promising framework for embodied intelligence, yet mainstream approaches remain centered on video foundation models or vision-language models. Even when geometric features extracted by 3D models such as VGGT are incorporated, they are typically passed into a video model or VLM that serves as the primary backbone, leaving 3D geometry as an auxiliary input or prediction target rather than the native foundation for world modeling. We present E0.5, a geometry-native world-action model built directly on 3D foundation models. E0.5 introduces two key innovations. First, it adopts a 3D foundation model as the primary backbone, making 3D geometry the native representation for both world modeling and action prediction. Second, it explicitly predicts future 3D states together with actions, enabling the model to capture geometric world dynamics beyond action-only modeling or future video prediction. Experiments on RoboTwin 2.0 first demonstrate the effectiveness of E0.5 without embodied-data pretraining, achieving success rates of 89.90\% and 50.09\% under clean and randomized settings, respectively. Building on these results, we further pretrain E0.5 on more than 160k robot trajectories from AgiBotWorld-Beta, jointly learning robot actions and future geometry within the pretrained 3D foundation. This embodied pretraining improves mean real-world manipulation success rates by approximately 15 and 7 percentage points on AgileX Piper and Zerith H1, respectively. Together, these results demonstrate the potential of geometry-native foundations to acquire robot interaction knowledge and transfer it to physical manipulation. E0.5 embodies this geometry-native approach as a member of the VWA family of world-action models.

cs.RO↗

arXiv2026

Embodied AI for Scientific Discovery: From Physical Action to Scientific Evidence

Scientific experiments are increasingly selected, executed and interpreted by learned systems that act through physical interfaces. Robot learning builds physical competence, whereas AI for science chooses which experiment to run, each reporting different endpoints. This review treats their interface as one problem, embodied AI for science, where the experiment itself is partially observed: specimen state, apparatus state and measurement processes are latent, actions are costly and irreversible, and missingness depends on execution and outcome. We organize the literature by three closures operating at distinct levels: operational closure (the experiment realized is the one intended), epistemic closure (the result updates beliefs or selects subsequent experiments) and evidential closure (the record supports a claim). We analyze workflows through a record-linked evidence cycle - Record, Diagnose, Calibrate, Confirm - as analytic functions with feedback. A taxonomy crosses these closures with six delegated decisions and five embodiment classes. Our formulation separates commanded, realized and observed conditions and models diagnosis with explicit resource, risk and sample-change costs. We compare experiment-selection methods by terminal decision, review perception, protocol compilation, visuomotor policies, world models, recovery and transfer by evidence emission requirements, and compare chemistry, materials, biology, microscopy and field workflows by what their records establish. An appendix provides an evidence map of 28 workflows (26 physical, two software-boundary), characterizing validation pathways and claim confirmation requirements. We propose evaluation that separates actions, valid experiments, campaigns and confirmed claims, framing open challenges as controlled comparisons with measurable failure conditions.

cs.AI↗

arXiv2019

Cleaning tasks knowledge transfer between heterogeneous robots: a deep learning approach

Nowadays, autonomous service robots are becoming an important topic in robotic research. Differently from typical industrial scenarios, with highly controlled environments, service robots must show an additional robustness to task perturbations and changes in the characteristics of their sensory feedback. In this paper, a robot is taught to perform two different cleaning tasks over a table, using a learning from demonstration paradigm. However, differently from other approaches, a convolutional neural network is used to generalize the demonstrations to different, not yet seen dirt or stain patterns on the same table using only visual feedback, and to perform cleaning movements accordingly. Robustness to robot posture and illumination changes is achieved using data augmentation techniques and camera images transformation. This robustness allows the transfer of knowledge regarding execution of cleaning tasks between heterogeneous robots operating in different environmental settings. To demonstrate the viability of the proposed approach, a network trained in Lisbon to perform cleaning tasks, using the iCub robot, is successfully employed by the DoRo robot in Peccioli, Italy.

cs.RO↗

arXiv2024

Optimization of Trajectories for Machine Learning Training in Robot Accuracy Modeling

Recently, machine learning (ML) methods have been developed for increasing the accuracy of robot mechanisms. Complex mechanical issues such as non-linear friction, backlash, flexibility of structure transmission elements can cause these errors and they are hard to model. ML requires training data and the above mechanical phenomena are highly dependent on position of the robot in the workspace and also on its velocity, especially near zero velocity in both directions where non-linearities such as Streibek and Coulomb friction are most pronounced. It is well known that success of ML methods depends on amount of training data and it is expensive/time consuming to collect data from physical robot motion. We therefore address the problem of searching for trajectories in the 6D space of positions and velocities which collect the most information in the least amount of time. This reduces to a special case of the traveling-salesman problem in that the robot must be programmed to visit sampled points in the position-velocity phase space most efficiently. Two goals of this work are 1) Computationally study the difficulty of the TSP in this application by applying it to X, Y, Z motion in 3D space (6D phase space) and 2) assess the effectiveness of an extremely simple Nearest Neighbor search algorithm compared to random sampling of the search space. Results confirm that Nearest Neighbor heuristic searching produces significantly better trajectories than random sampling in this application.

cs.RO↗