Stéphane Doncieux

Aurélien Morel, Yakumo Kunimoto, Alex Coninx et al.

Grasping a particular object may require a dedicated grasping movement that may also be specific to the robot end-effector. No generic and autonomous method does exist to generate these movements without making hypotheses on the robot or on the object. Learning methods could help to autonomously discover relevant grasping movements, but they face an important issue: grasping movements are so rare that a learning method based on exploration has little chance to ever observe an interesting movement, thus creating a bootstrap issue. We introduce an approach to generate diverse grasping movements in order to solve this problem. The movements are generated in simulation, for particular object positions. We test it on several simulated robots: Baxter, Pepper and a Kuka Iiwa arm. Although we show that generated movements actually work on a real Baxter robot, the aim is to use this method to create a large dataset to bootstrap deep learning methods.

Johann Huber, François Hélénon, Hippolyte Watrelot et al.

Robotic grasping refers to making a robotic system pick an object by applying forces and torques on its surface. Many recent studies use data-driven approaches to address grasping, but the sparse reward nature of this task made the learning process challenging to bootstrap. To avoid constraining the operational space, an increasing number of works propose grasping datasets to learn from. But most of them are limited to simulations. The present paper investigates how automatically generated grasps can be exploited in the real world. More than 7000 reach-and-grasp trajectories have been generated with Quality-Diversity (QD) methods on 3 different arms and grippers, including parallel fingers and a dexterous hand, and tested in the real world. Conducted analysis on the collected measure shows correlations between several Domain Randomization-based quality criteria and sim-to-real transferability. Key challenges regarding the reality gap for grasping have been identified, stressing matters on which researchers on grasping should focus in the future. A QD approach has finally been proposed for making grasps more robust to domain randomization, resulting in a transfer ratio of 84% on the Franka Research 3 arm.

Mathilde Kappel, Mahdi Khoramshahi, Louis Annabi et al.

Thanks to the latest advances in learning and robotics, domestic robots are beginning to enter homes, aiming to execute household chores autonomously. However, robots still struggle to perform autonomous manipulation tasks in open-ended environments. In this context, this paper presents a method that enables a robot to manipulate a wide spectrum of articulated objects. In this paper, we automatically generate different robot low-level trajectory primitives to manipulate given object articulations. A very important point when it comes to generating expert trajectories is to consider the diversity of solutions to achieve the same goal. Indeed, knowing diverse low-level primitives to accomplish the same task enables the robot to choose the optimal solution in its real-world environment, with live constraints and unexpected changes. To do so, we propose a method based on Quality-Diversity algorithms that leverages sparse reward exploration in order to generate a set of diverse and high-performing trajectory primitives for a given manipulation task. We validated our method, QDTraj, by generating diverse trajectories in simulation and deploying them in the real world. QDTraj generates at least 5 times more diverse trajectories for both hinge and slider activation tasks, outperforming the other methods we compared against. We assessed the generalization of our method over 30 articulations of the PartNetMobility articulated object dataset, with an average of 704 different trajectories by task. Code is publicly available at: https://kappel.web.isir.upmc.fr/trajectory_primitive_website

François Hélénon, Johann Huber, Faïz Ben Amar et al.

Despite recent advancements in AI for robotics, grasping remains a partially solved challenge, hindered by the lack of benchmarks and reproducibility constraints. This paper introduces a vision-based grasping framework that can easily be transferred across multiple manipulators. Leveraging Quality-Diversity (QD) algorithms, the framework generates diverse repertoires of open-loop grasping trajectories, enhancing adaptability while maintaining a diversity of grasps. This framework addresses two main issues: the lack of an off-the-shelf vision module for detecting object pose and the generalization of QD trajectories to the whole robot operational space. The proposed solution combines multiple vision modules for 6DoF object detection and tracking while rigidly transforming QD-generated trajectories into the object frame. Experiments on a Franka Research 3 arm and a UR5 arm with a SIH Schunk hand demonstrate comparable performance when the real scene aligns with the simulation used for grasp generation. This work represents a significant stride toward building a reliable vision-based grasping module transferable to new platforms, while being adaptable to diverse scenarios without further training iterations.

Elias Hanna, Alex Coninx, Stéphane Doncieux

This paper studies the impact of the initial data gathering method on the subsequent learning of a dynamics model. Dynamics models approximate the true transition function of a given task, in order to perform policy search directly on the model rather than on the costly real system. This study aims to determine how to bootstrap a model as efficiently as possible, by comparing initialization methods employed in two different policy search frameworks in the literature. The study focuses on the model performance under the episode-based framework of Evolutionary methods using probabilistic ensembles. Experimental results show that various task-dependant factors can be detrimental to each method, suggesting to explore hybrid approaches.

Johann Huber, François Hélénon, Mathilde Kappel et al.

Recent advances in AI have led to significant results in robotic learning, including natural language-conditioned planning and efficient optimization of controllers using generative models. However, the interaction data remains the bottleneck for generalization. Getting data for grasping is a critical challenge, as this skill is required to complete many manipulation tasks. Quality-Diversity (QD) algorithms optimize a set of solutions to get diverse, high-performing solutions to a given problem. This paper investigates how QD can be combined with priors to speed up the generation of diverse grasps poses in simulation compared to standard 6-DoF grasp sampling schemes. Experiments conducted on 4 grippers with 2-to-5 fingers on standard objects show that QD outperforms commonly used methods by a large margin. Further experiments show that QD optimization automatically finds some efficient priors that are usually hard coded. The deployment of generated grasps on a 2-finger gripper and an Allegro hand shows that the diversity produced maintains sim-to-real transferability. We believe these results to be a significant step toward the generation of large datasets that can lead to robust and generalizing robotic grasping policies.

Amina Ferrad, Johann Huber, François Hélénon et al.

Robotics research has made significant strides in learning, yet mastering basic skills like object placement remains a fundamental challenge. A key bottleneck is the acquisition of large-scale, high-quality data, which is often a manual and laborious process. Inspired by Graspit!, a foundational work that used simulation to automatically generate dexterous grasp poses, we introduce Placeit!, an evolutionary-computation framework for generating valid placement positions for rigid objects. Placeit! is highly versatile, supporting tasks from placing objects on tables to stacking and inserting them. Our experiments show that by leveraging quality-diversity optimization, Placeit! significantly outperforms state-of-the-art methods across all scenarios for generating diverse valid poses. A pick&place pipeline built on our framework achieved a 90% success rate over 120 real-world deployments. This work positions Placeit! as a powerful tool for open-environment pick-and-place tasks and as a valuable engine for generating the data needed to train simulation-based foundation models in robotics.

Olivier Serris, Stéphane Doncieux, Olivier Sigaud

Several hierarchical reinforcement learning methods leverage planning to create a graph or sequences of intermediate goals, guiding a lower-level goal-conditioned (GC) policy to reach some final goals. The low-level policy is typically conditioned on the current goal, with the aim of reaching it as quickly as possible. However, this approach can fail when an intermediate goal can be reached in multiple ways, some of which may make it impossible to continue toward subsequent goals. To address this issue, we introduce two instances of Markov Decision Process (MDP) where the optimization objective favors policies that not only reach the current goal but also subsequent ones. In the first, the agent is conditioned on both the current and final goals, while in the second, it is conditioned on the next two goals in the sequence. We conduct a series of experiments on navigation and pole-balancing tasks in which sequences of intermediate goals are given. By evaluating policies trained with TD3+HER on both the standard GC-MDP and our proposed MDPs, we show that, in most cases, conditioning on the next two goals improves stability and sample efficiency over other approaches.

Charly Pecqueux-Guézénec, Stéphane Doncieux, Nicolas Perrin-Gilbert

Goal-Conditioned Reinforcement Learning (GCRL) provides a versatile framework for developing unified controllers capable of handling wide ranges of tasks, exploring environments, and adapting behaviors. However, its reliance on trial-and-error poses challenges for real-world applications, as errors can result in costly and potentially damaging consequences. To address the need for safer learning, we propose a method that enables agents to learn goal-conditioned behaviors that explore without the risk of making harmful mistakes. Exploration without risks can seem paradoxical, but environment dynamics are often uniform in space, therefore a policy trained for safety without exploration purposes can still be exploited globally. Our proposed approach involves two distinct phases. First, during a pretraining phase, we employ safe reinforcement learning and distributional techniques to train a safety policy that actively tries to avoid failures in various situations. In the subsequent safe exploration phase, a goal-conditioned (GC) policy is learned while ensuring safety. To achieve this, we implement an action-selection mechanism leveraging the previously learned distributional safety critics to arbitrate between the safety policy and the GC policy, ensuring safe exploration by switching to the safety policy when needed. We evaluate our method in simulated environments and demonstrate that it not only provides substantial coverage of the goal space but also reduces the occurrence of mistakes to a minimum, in stark contrast to traditional GCRL approaches. Additionally, we conduct an ablation study and analyze failure modes, offering insights for future research directions.

Astrid Merckling, Nicolas Perrin-Gilbert, Alex Coninx et al.

Not having access to compact and meaningful representations is known to significantly increase the complexity of reinforcement learning (RL). For this reason, it can be useful to perform state representation learning (SRL) before tackling RL tasks. However, obtaining a good state representation can only be done if a large diversity of transitions is observed, which can require a difficult exploration, especially if the environment is initially reward-free. To solve the problems of exploration and SRL in parallel, we propose a new approach called XSRL (eXploratory State Representation Learning). On one hand, it jointly learns compact state representations and a state transition estimator which is used to remove unexploitable information from the representations. On the other hand, it continuously trains an inverse model, and adds to the prediction error of this model a $k$-step learning progress bonus to form the maximization objective of a discovery policy. This results in a policy that seeks complex transitions from which the trained models can effectively learn. Our experimental results show that the approach leads to efficient exploration in challenging environments with image observations, and to state representations that significantly accelerate learning in RL tasks.

Alexandre Chenu, Nicolas Perrin-Gilbert, Stéphane Doncieux et al.

Reinforcement learning agents need a reward signal to learn successful policies. When this signal is sparse or the corresponding gradient is deceptive, such agents need a dedicated mechanism to efficiently explore their search space without relying on the reward. Looking for a large diversity of behaviors or using Motion Planning (MP) algorithms are two options in this context. In this paper, we build on the common roots between these two options to investigate the properties of two diversity search algorithms, the Novelty Search and the Goal Exploration Process algorithms. These algorithms look for diversity in an outcome space or behavioral space which is generally hand-designed to represent what matters for a given task. The relation to MP algorithms reveals that the smoothness, or lack of smoothness of the mapping between the policy parameter space and the outcome space plays a key role in the search efficiency. In particular, we show empirically that, if the mapping is smooth enough, i.e. if two close policies in the parameter space lead to similar outcomes, then diversity algorithms tend to inherit exploration properties of MP algorithms. By contrast, if it is not, diversity algorithms lose these properties and their performance strongly depends on specific heuristics, notably filtering mechanisms that discard some of the explored policies.

Astrid Merckling, Alexandre Coninx, Loic Cressot et al.

Robots could learn their own state and world representation from perception and experience without supervision. This desirable goal is the main focus of our field of interest, state representation learning (SRL). Indeed, a compact representation of such a state is beneficial to help robots grasp onto their environment for interacting. The properties of this representation have a strong impact on the adaptive capability of the agent. In this article we present an approach based on imitation learning. The idea is to train several policies that share the same representation to reproduce various demonstrations. To do so, we use a multi-head neural network with a shared state representation feeding a task-specific agent. If the demonstrations are diverse, the trained representation will eventually contain the information necessary for all tasks, while discarding irrelevant information. As such, it will potentially become a compact state representation useful for new tasks. We call this approach SRLfD (State Representation Learning from Demonstration). Our experiments confirm that when a controller takes SRLfD-based representations as input, it can achieve better performance than with other representation strategies and promote more efficient reinforcement learning (RL) than with an end-to-end RL strategy.

Léni K. Le Goff, Ghanim Mukhtar, Alexandre Coninx et al.

To solve its task, a robot needs to have the ability to interpret its perceptions. In vision, this interpretation is particularly difficult and relies on the understanding of the structure of the scene, at least to the extent of its task and sensorimotor abilities. A robot with the ability to build and adapt this interpretation process according to its own tasks and capabilities would push away the limits of what robots can achieve in a non controlled environment. A solution is to provide the robot with processes to build such representations that are not specific to an environment or a situation. A lot of works focus on objects segmentation, recognition and manipulation. Defining an object solely on the basis of its visual appearance is challenging given the wide range of possible objects and environments. Therefore, current works make simplifying assumptions about the structure of a scene. Such assumptions reduce the adaptivity of the object extraction process to the environments in which the assumption holds. To limit such assumptions, we introduce an exploration method aimed at identifying moveable elements in a scene without considering the concept of object. By using the interactive perception framework, we aim at bootstrapping the acquisition process of a representation of the environment with a minimum of context specific assumptions. The robotic system builds a perceptual map called relevance map which indicates the moveable parts of the current scene. A classifier is trained online to predict the category of each region (moveable or non-moveable). It is also used to select a region with which to interact, with the goal of minimizing the uncertainty of the classification. A specific classifier is introduced to fit these needs: the collaborative mixture models classifier. The method is tested on a set of scenarios of increasing complexity, using both simulations and a PR2 robot.

Marija Jegorova, Stéphane Doncieux, Timothy Hospedales

Learning algorithms are enabling robots to solve increasingly challenging real-world tasks. These approaches often rely on demonstrations and reproduce the behavior shown. Unexpected changes in the environment may require using different behaviors to achieve the same effect, for instance to reach and grasp an object in changing clutter. An emerging paradigm addressing this robustness issue is to learn a diverse set of successful behaviors for a given task, from which a robot can select the most suitable policy when faced with a new environment. In this paper, we explore a novel realization of this vision by learning a generative model over policies. Rather than learning a single policy, or a small fixed repertoire, our generative model for policies compactly encodes an unbounded number of policies and allows novel controller variants to be sampled. Leveraging our generative policy network, a robot can sample novel behaviors until it finds one that works for a new environment. We demonstrate this idea with an application of robust ball-throwing in the presence of obstacles. We show that this approach achieves a greater diversity of behaviors than an existing evolutionary approach, while maintaining good efficacy of sampled behaviors, allowing a Baxter robot to hit targets more often when ball throwing in the presence of obstacles.

Stéphane Doncieux, Jean Liénard, Benoît Girard et al.

Computational models are of increasing complexity and their behavior may in particular emerge from the interaction of different parts. Studying such models becomes then more and more difficult and there is a need for methods and tools supporting this process. Multi-objective evolutionary algorithms generate a set of trade-off solutions instead of a single optimal solution. The availability of a set of solutions that have the specificity to be optimal relative to carefully chosen objectives allows to perform data mining in order to better understand model features and regularities. We review the corresponding work, propose a unifying framework, and highlight its potential use. Typical questions that such a methodology allows to address are the following: what are the most critical parameters of the model? What are the relations between the parameters and the objectives? What are the typical behaviors of the model? Two examples are provided to illustrate the capabilities of the methodology. The features of a flapping-wing robot are thus evaluated to find out its speed-energy relation, together with the criticality of its parameters. A neurocomputational model of the Basal Ganglia brain nuclei is then considered and its most salient features according to this methodology are presented and discussed.

Jean-Baptiste Mouret, Sylvain Koos, Stéphane Doncieux

In robotics, gradient-free optimization algorithms (e.g. evolutionary algorithms) are often used only in simulation because they require the evaluation of many candidate solutions. Nevertheless, solutions obtained in simulation often do not work well on the real device. The transferability approach aims at crossing this gap between simulation and reality by \emph{making the optimization algorithm aware of the limits of the simulation}. In the present paper, we first describe the transferability function, that maps solution descriptors to a score representing how well a simulator matches the reality. We then show that this function can be learned using a regression algorithm and a few experiments with the real devices. Our results are supported by an extensive study of the reality gap for a simple quadruped robot whose control parameters are optimized. In particular, we mapped the whole search space in reality and in simulation to understand the differences between the fitness landscapes.