Robotic manipulation faces a significant challenge in generalizing across unseen objects, environments and tasks specified by diverse language instructions. To improve generalization capabilities, recent research has incorporated large language models (LLMs) for planning and action execution. While promising, these methods often fall short in generating grounded plans in visual environments. Although efforts have been made to perform visual instructional tuning on LLMs for robotic manipulation, existing methods are typically constrained by single-view image input and struggle with precise object grounding. In this work, we introduce Gondola, a novel grounded vision-language planning model based on LLMs for generalizable robotic manipulation. Gondola takes multi-view images and history plans to produce the next action plan with interleaved texts and segmentation masks of target objects and locations. To support the training of Gondola, we construct three types of datasets using the RLBench simulator, namely robot grounded planning, multi-view referring expression and pseudo long-horizon task datasets. Gondola outperforms the state-of-the-art LLM-based method across all four generalization levels of the GemBench dataset, including novel placements, rigid objects, articulated objects and long-horizon tasks.

Deep reinforcement learning (RL) agents achieve impressive results in a wide variety of tasks, but they lack zero-shot adaptation capabilities. While most robustness evaluations focus on tasks complexifications, for which human also struggle to maintain performances, no evaluation has been performed on tasks simplifications. To tackle this issue, we introduce HackAtari, a set of task variations of the Arcade Learning Environments. We use it to demonstrate that, contrary to humans, RL agents systematically exhibit huge performance drops on simpler versions of their training tasks, uncovering agents' consistent reliance on shortcuts. Our analysis across multiple algorithms and architectures highlights the persistent gap between RL agents and human behavioral intelligence, underscoring the need for new benchmarks and methodologies that enforce systematic generalization testing beyond static evaluation protocols. Training and testing in the same environment is not enough to obtain agents equipped with human-like intelligence.

When using Large Language Models (LLMs) to support Knowledge Graph Engineering (KGE), one of the first indications when searching for an appropriate model is its size. According to the scaling laws, larger models typically show higher capabilities. However, in practice, resource costs are also an important factor and thus it makes sense to consider the ratio between model performance and costs. The LLM-KG-Bench framework enables the comparison of LLMs in the context of KGE tasks and assesses their capabilities of understanding and producing KGs and KG queries. Based on a dataset created in an LLM-KG-Bench run covering 26 open state-of-the-art LLMs, we explore the model size scaling laws specific to KGE tasks. In our analyses, we assess how benchmark scores evolve between different model size categories. Additionally, we inspect how the general score development of single models and families of models correlates to their size. Our analyses revealed that, with a few exceptions, the model size scaling laws generally also apply to the selected KGE tasks. However, in some cases, plateau or ceiling effects occurred, i.e., the task performance did not change much between a model and the next larger model. In these cases, smaller models could be considered to achieve high cost-effectiveness. Regarding models of the same family, sometimes larger models performed worse than smaller models of the same family. These effects occurred only locally. Hence it is advisable to additionally test the next smallest and largest model of the same family.

With the rising demand for flexible manufacturing, robots are increasingly expected to operate in dynamic environments where local disturbances--such as slight offsets or size differences in workpieces--are common. We propose to address the problem of adapting robot behaviors to these task variations with a sample-efficient hierarchical reinforcement learning approach adapting Behavior Tree (BT)-based policies. We maintain the core BT properties as an interpretable, modular framework for structuring reactive behaviors, but extend their use beyond static tasks by inherently accommodating local task variations. To show the efficiency and effectiveness of our approach, we conduct experiments both in simulation and on a Franka Emika Panda 7-DoF, with the manipulator adapting to different obstacle avoidance and pivoting tasks.

This paper primarily focuses on learning robust visual-force policies in the context of high-precision object assembly tasks. Specifically, we focus on the contact phase of the assembly task where both objects (peg and hole) have made contact and the objective lies in maneuvering the objects to complete the assembly. Moreover, we aim to learn contact-rich manipulation policies with multisensory inputs on limited expert data by expanding human demonstrations via online data augmentation. We develop a simulation environment with a dual-arm robot manipulator to evaluate the effect of augmented expert demonstration data. Our focus is on evaluating the robustness of our model with respect to certain task variations: grasp pose, peg/hole shape, object body shape, scene appearance, camera pose, and force-torque/proprioception noise. We show that our proposed data augmentation method helps in learning a multisensory manipulation policy that is robust to unseen instances of these variations, particularly physical variations such as grasp pose. Additionally, our ablative studies show the significant contribution of force-torque data to the robustness of our model. For additional experiments and qualitative results, we refer to the project webpage at https://bit.ly/47skWXH .

Visual Reinforcement Learning is a popular and powerful framework that takes full advantage of the Deep Learning breakthrough. However, it is also known that variations in the input (e.g., different colors of the panorama due to the season of the year) or the task (e.g., changing the speed limit for a car to respect) could require complete retraining of the agents. In this work, we leverage recent developments in unifying latent representations to demonstrate that it is possible to combine the components of an agent, rather than retrain it from scratch. We build upon the recent relative representations framework and adapt it for Visual RL. This allows us to create completely new agents capable of handling environment-task combinations never seen during training. Our work paves the road toward a more accessible and flexible use of reinforcement learning.

This paper describes a novel approach to adaptive manufacturing in the context of small batch production and customization. It focuses on integrating task-level planning and reasoning with reinforcement learning (RL) in the SkiROS2 skill-based robot control platform. This integration enhances the efficiency and adaptability of robotic systems in manufacturing, enabling them to adjust to task variations and learn from interaction data. The paper highlights the architecture of SkiROS2, particularly its world model, skill libraries, and task management. It demonstrates how combining RL with robotic manipulators can learn and improve the execution of industrial tasks. It advocates a multi-objective learning model that eases the learning problem design. The approach can incorporate user priors or previous experiences to accelerate learning and increase safety. Spotlight video: https://youtu.be/H5PmZl2rRbs?si=8wmZ-gbwuSJRxe3S&t=1422 SkiROS2 code: https://github.com/RVMI/skiros2 SkiROS2 talk at ROSCon: https://vimeo.com/879001825/2a0e9d5412 SkiREIL code: https://github.com/matthias-mayr/SkiREIL

The ability to learn new tasks and quickly adapt to different variations or dimensions is an important attribute in agile robotics. In our previous work, we have explored Behavior Trees and Motion Generators (BTMGs) as a robot arm policy representation to facilitate the learning and execution of assembly tasks. The current implementation of the BTMGs for a specific task may not be robust to the changes in the environment and may not generalize well to different variations of tasks. We propose to extend the BTMG policy representation with a module that predicts BTMG parameters for a new task variation. To achieve this, we propose a model that combines a Gaussian process and a weighted support vector machine classifier. This model predicts the performance measure and the feasibility of the predicted policy with BTMG parameters and task variations as inputs. Using the outputs of the model, we then construct a surrogate reward function that is utilized within an optimizer to maximize the performance of a task over BTMG parameters for a fixed task variation. To demonstrate the effectiveness of our proposed approach, we conducted experimental evaluations on push and obstacle avoidance tasks in simulation and with a real KUKA iiwa robot. Furthermore, we compared the performance of our approach with four baseline methods.

Tasks where the set of possible actions depend discontinuously on the state pose a significant challenge for current reinforcement learning algorithms. For example, a locked door must be first unlocked, and then the handle turned before the door can be opened. The sequential nature of these tasks makes obtaining final rewards difficult, and transferring information between task variants using continuous learned values such as weights rather than discrete symbols can be inefficient. Our key insight is that agents that act and think symbolically are often more effective in dealing with these tasks. We propose a memory-based learning approach that leverages the symbolic nature of constraints and temporal ordering of actions in these tasks to quickly acquire and transfer high-level information. We evaluate the performance of memory-based learning on both real and simulated tasks with approximately discontinuous constraints between states and actions, and show our method learns to solve these tasks an order of magnitude faster than both model-based and model-free deep reinforcement learning methods.