IROS 2024 Papers

We propose a novel framework and algorithm for solving classical planning problems with an implicit hierarchical solver based on the principle of delegation. This framework, the Markov Intent Process, features a collection of skills that are each specialised to perform a single task well. Skills are aware of their intended effects and are able to analyse planning goals to delegate planning to the best-suited skill. This principle dynamically creates a hierarchy of plans, in which each skill plans for sub-goals for which it is specialised. Our method performs robustly in noisy environments with non-deterministic action effects and features on-demand execution—skill policies are only evaluated when needed. Plans are only generated at the highest level, then expanded and optimised when the latest state information is available. The high-level plan retains the initial planning intent and previously computed skills, effectively reducing the computation needed to adapt to environmental changes. We show this planning approach is experimentally very competitive to classic planning and reinforcement learning techniques on a variety of domains, both in terms of solution length and planning time.

While robust model predictive control (MPC) has been studied extensively in recent decades, addressing unmatched disturbances in underactuated robotic systems is still challenging. In this paper, we propose a method to enhance the robustness of the MPC through the online estimation of disturbances using a nonlinear disturbance observer (NDOB). We call this method disturbance-aware MPC (DA-MPC), because the proposed method explicitly utilizes the estimated disturbance in the future prediction. We provide a performance analysis of the NDOB, establishing the boundedness between the predicted and real states. The main advantages of the DA-MPC include its applicability to real-time control and its compatibility with off-the-shelf optimal control problem (OCP) solvers. We demonstrate the application of the proposed method using an underactuated quadrotor system. The simulation validation shows the effectiveness of the proposed method compared to ℒ1-adaptive MPC, which is one of the state-of-the-art robust MPC methods.

This paper proposes Distributed Model Predictive Covariance Steering (DiMPCS) for multi-agent control under stochastic uncertainty. The scope of our approach is to blend covariance steering theory, distributed optimization and model predictive control (MPC) into a single framework that is safe, scalable and decentralized. Initially, we pose a problem formulation that uses the Wasserstein distance to steer the state distributions of a multi-agent system to desired targets, and probabilistic constraints to ensure safety. We then transform this problem into a finite-dimensional optimization one by utilizing a disturbance feedback policy parametrization for covariance steering and a tractable approximation of the safety constraints. To solve the latter problem, we derive a decentralized consensus-based algorithm using the Alternating Direction Method of Multipliers. This method is then extended to a receding horizon form, which yields the proposed DiMPCS algorithm. Simulation experiments on a variety of multi-robot tasks with up to hundreds of robots demonstrate the effectiveness of DiMPCS. The superior scalability and performance of the proposed method is also highlighted through a comparison against related stochastic MPC approaches. Finally, hardware results on a multi-robot platform also verify the applicability of DiMPCS on real systems.

Recent advancements in reinforcement learning (RL) have led to remarkable achievements in robot locomotion capabilities. However, the complexity and "black-box" nature of neural network-based RL policies hinder their interpretability and broader acceptance, particularly in applications demanding high levels of safety and reliability. This paper introduces a novel approach to distill neural RL policies into more interpretable forms using Gradient Boosting Machines (GBMs), Explainable Boosting Machines (EBMs) and Symbolic Regression. By leveraging the inherent interpretability of generalized additive models, decision trees, and analytical expressions, we transform opaque neural network policies into more transparent "glass-box" models. We train expert neural network policies using RL and subsequently distill them into (i) GBMs, (ii) EBMs, and (iii) symbolic policies. To address the inherent distribution shift challenge of behavioral cloning, we propose to use the Dataset Aggregation (DAgger) algorithm with a curriculum of episode-dependent alternation of actions between expert and distilled policies, to enable efficient distillation of feedback control policies. We evaluate our approach on various robot locomotion gaits – walking, trotting, bounding, and pacing – and study the importance of different observations in joint actions for distilled policies using various methods. We train neural expert policies for 205 hours of simulated experience and distill interpretable policies with only 10 minutes of simulated interaction for each gait using the proposed method.

We study the problem of multimodal physical scene understanding, where an embodied agent needs to find fallen objects by inferring object properties, direction, and distance of an impact sound source. Previous works adopt feed-forward neural networks to directly regress the variables from sound, leading to poor generalization and domain adaptation issues. In this paper, we illustrate that learning a disentangled model of acoustic formation, referred to as disentangled acoustic field (DAF), to capture the sound generation and propagation process, enables the embodied agent to construct a spatial uncertainty map over where the objects may have fallen. We demonstrate that our analysis-by-synthesis framework can jointly infer sound properties by explicitly decomposing and factorizing the latent space of the disentangled model. We further show that the spatial uncertainty map can significantly improve the success rate for the localization of fallen objects by proposing multiple plausible exploration locations.

Robotic kitting has attracted considerable attention in logistics and industrial settings. However, existing kitting methods encounter challenges such as low precision and poor efficiency, limiting their widespread applications. To address these issues, we present a novel kitting framework that improves both the precision and computational efficiency of complex kitting tasks. Firstly, our approach introduces a fine-grained orientation estimation technique in the picking module, significantly enhancing orientation precision while effectively decoupling computational load from orientation granularity. This technique combines an SO(2)-equivariant network with a group discretization operation to preciously predict discrete orientation distributions. Secondly, we develop the Hand-Tool Kitting Dataset (HTKD) to evaluate different solutions in handling orientation-sensitive kitting tasks. This dataset comprises a diverse collection of hand tools and synthetically created kits, which reflects the complexities of real-world kitting scenarios. Finally, a series of experiments is conducted to evaluate the performance of the proposed method. The results demonstrate that our approach offers an excellent balance between success rates and computational efficiency in high-precision robotic kitting tasks.

Recent advancements in gait rehabilitation have led to the development of innovative approaches that complement traditional therapeutic methods. These include intense, task-specific exercise strategies, non-invasive treatments, surgical interventions, and advanced robotic technologies. While robotic systems for adult gait rehabilitation are well-established, there is a notable scarcity of such devices for pediatric patients, especially toddlers, due to their unique developmental and biomechanical needs. This work introduces Discover2Walk (D2W), a novel robotic platform designed specifically for pediatric gait rehabilitation in small children. The D2W platform features a multi-module, cable-driven architecture guided by an omnidirectional traction module, addressing the limitations of current rehabilitation devices for younger populations. The platform's modular design consists of three actuated modules—pelvic, ankle, and traction—synchronized by a personalised gait pattern generator. This configuration allows for simultaneous control and monitoring of pelvic and ankle motion using partial body weight support and Assistance As Needed (AAN) approaches. Preliminary evaluations were conducted with pediatric patients with Cerebral Palsy, involving two ambulatory six-year-olds and one non-ambulatory four-year-old, over a series of 10 gait rehabilitation sessions. Data analysis from the ambulatory children showed a decrease in the robotic effort required to assist limb movements along healthy trajectories throughout the sessions, accompanied by an increase in walking speeds. Further work will include expanding the patient cohort to include a broader range of ages, sizes and GMFCS levels to validate the system's effectiveness across a wider spectrum of pediatric gait disabilities and validating the traction effectiveness.

In this paper, a method for 3D non-rigid motion estimation of a surface using an RGB-D camera in eye-in-hand configuration is presented. The eye-in-hand configuration eliminates errors typically associated with camera-end-effector calibration, and is thus desirable for task on moving surfaces such as bioprinting. However, its implementation is challenging since camera and surface of interest are moving, making mesh-based approaches unsuitable. Thus, the proposed method operates directly on point clouds, benefiting from accurate and simplified data processing. A point cloud contains both intensity and depth data, with the former used to estimate in-plane deformation and the latter to compute full 3D deformation. Surface deformation is modeled via a Thin Plate Spline model. The method accuracy is assessed at 0.1 mm accuracy in simulated datasets, rendering it suitable for precision tasks, and its feasibility is validated experimentally on a moving platform that deforms at a rate of 0.8 Hz with a 4 mm in-plane amplitude and a 20 mm elevation amplitude.

This paper addresses the challenging problem of category-level pose estimation. Current state-of-the-art methods for this task face challenges when dealing with symmetric objects and when attempting to generalize to new environments solely through synthetic data training. In this work, we address these challenges by proposing a probabilistic model that relies on diffusion to estimate dense canonical maps crucial for recovering partial object shapes as well as establishing correspondences essential for pose estimation. Furthermore, we introduce critical components to enhance performance by leveraging the strength of the diffusion models with multi-modal input representations. We demonstrate the effectiveness of our method by testing it on a range of real datasets. Despite being trained solely on our generated synthetic data, our approach achieves state-of-the-art performance and unprecedented generalization qualities, outperforming baselines, even those specifically trained on the target domain. Our code and data for the generalization benchmark can be found at https://woven-planet.github.io/DiffusionNOCS/.

Programming by Demonstration (PbD) is an intuitive technique for programming robot manipulation skills by demonstrating the desired behavior. However, most existing approaches either require extensive demonstrations or fail to generalize beyond their initial demonstration conditions. We introduce Diffusion-PbD, a novel approach to PbD that enables users to synthesize generalizable robot manipulation skills from a single demonstration by utilizing the representations captured by pre-trained visual foundation models. At demonstration time, hand and object detection priors are used to extract waypoints from the human demonstrations anchored to reference points in the scene. At execution time, features from pre-trained diffusion models are leveraged to identify corresponding reference points in new observations. We validate this approach through a series of real-world robot experiments, showing that Diffusion-PbD is applicable to a wide range of manipulation tasks and has strong ability to generalize to unseen objects, camera viewpoints, and scenes. Code and supplementary videos can be found at https://diffusion-pbd.github.io

This paper presents a control barrier function (CBF) for systems described by differential-algebraic equations and applies the method to guarantee the safety of a two-link flexible-link manipulator. The two main contributions of the paper are: a) an extension of CBFs to systems governed by differential-algebraic equations; b) a framework for simulation of flexible-link robots in a floating frame of reference formulation (FFRF) finite element method (FEM). Numerical simulations demonstrate the minimally invasive safety control of a flexible two-link manipulator with position constraints through CBF quadratic programming without converting the differential-algebraic equations to a control-affine system.

Fluid interactions are crucial in daily tasks, with properties like density and viscosity being key parameters. The property states can be used as control signals for robot operation. While density estimation is simple, assessing viscosity, especially for different fluid types, is complex. This study introduces a novel differentiable fitting framework, DiffStir, tailored to identify key physics parameters through stirring. Then, given the estimated physics parameters, we can generate commands to guide the robotic stirring. Comprehensive experiments were conducted to validate the efficacy of DiffStir, showcasing its precision in parameter estimation when benchmarked against reported values in the literature. More experiments and videos can be found in the supplementary materials and on the website: https://diffstir.robotflow.ai.

This paper presents a method to compute differentiable collision-free parametric corridors. In contrast to existing solutions that decompose the obstacle-free space into multiple convex sets, the continuous corridors computed by our method are smooth and differentiable, making them compatible with existing numerical techniques for learning and optimization. To achieve this, we represent the collision-free corridors as a path-parametric off-centered ellipse with a polynomial basis. We show that the problem of maximizing the volume of such corridors is convex, and can be efficiently solved. To assess the effectiveness of the proposed method, we examine its performance in a synthetic case study and subsequently evaluate its applicability in a real-world scenario from the KITTI dataset.

Semantic segmentation in adverse weather scenarios is a critical task for autonomous driving systems. While foundation models have shown promise, the need for specialized adaptors becomes evident for handling more challenging scenarios. We introduce DiffPrompter, a novel differentiable visual and latent prompting mechanism aimed at expanding the learning capabilities of existing adaptors in foundation models. Our proposed ∇HFC (High Frequency Components) based image processing block excels particularly in adverse weather conditions, where conventional methods often fall short. Furthermore, we investigate the advantages of jointly training visual and latent prompts, demonstrating that this combined approach significantly enhances performance in out-of-distribution scenarios. Our differentiable visual prompts leverage parallel and series architectures to generate prompts, effectively improving object segmentation tasks in adverse conditions. Through a comprehensive series of experiments and evaluations, we provide empirical evidence to support the efficacy of our approach. Project page: diffprompter.github.io

While imitation learning provides a simple and effective framework for policy learning, acquiring consistent action during robot execution remains a challenging task. Existing approaches primarily focus on either modifying the action representation at data curation stage or altering the model itself, both of which do not fully address the scalability of consistent action generation. To overcome this limitation, we introduce the Diff-Control policy, which utilizes a diffusion-based model to learn action representation from a state-space modeling viewpoint. We demonstrate that diffusion-based policies can acquire statefulness through a Bayesian formulation facilitated by ControlNet, leading to improved robustness and success rates. Our experimental results demonstrate the significance of incorporating action statefulness in policy learning, where Diff-Control shows improved performance across various tasks. Specifically, Diff-Control achieves an average success rate of 72% and 84% on stateful and dynamic tasks, respectively. Notably, Diff-Control also shows consistent performance in the presence of perturbations, outperforming other state-of-the-art methods that falter under similar conditions. Project page: https://diff-control.github.io/

Manipulating robots using natural language is the preferred way for non-technical specialists. The challenge lies in reliability and adaptability especially when robots operate in unstructured surroundings. In this paper, we propose a novel framework called Dialogue Generative Behavior Trees (DiaGBT). Natural language instructions from human operators are transformed into behavior trees (BTs) and further executed by robots. Compared to the emerging Large Language Models (LLMs), DiaGBT is comparable in terms of semantic understanding but more lightweight, since the parsing rules are produced by LLM but tailored for task-correlated instructions. Besides, DiaGBT allows multi-round human-robot interaction, where robots learn reusable skills in real time. For evaluation, we generate a dataset with 4k instruction-BT pairs covering 4 different scenarios. On average, DiaGBT reaches over 90% parsability and 80% plausibility. Similar results on the VEIL-500 dataset outperform the current state of the art.

We present DiPPeST, a novel image and goal conditioned diffusion-based trajectory generator for quadrupedal robot path planning. DiPPeST is a zero-shot adaptation of our previously introduced diffusion-based 2D global trajectory generator (DiPPeR). The introduced system incorporates a novel strategy for local real-time path refinements, that is reactive to camera input, without requiring any further training, image processing, or environment interpretation techniques. DiPPeST achieves 92% success rate in obstacle avoidance for nominal environments and an average of 88% success rate when tested in environments that are up to 3.5 times more complex in pixel variation than DiPPeR. A visual-servoing framework is developed to allow for real-world execution, tested on the quadruped robot, achieving 80% success rate in different environments and showcasing improved behavior than complex state-of-the-art local planners, in narrow environments. Website: https://rpl-cs-ucl.github.io/DiPPeSTweb/

Generative models such as diffusion models, excel at capturing high-dimensional distributions with diverse input modalities, e.g. robot trajectories, but are less effective at multistep constraint reasoning. Task and Motion Planning (TAMP) approaches are suited for planning multi-step autonomous robot manipulation. However, it can be difficult to apply them to domains where the environment and its dynamics are not fully known. We propose to overcome these limitations by composing diffusion models using a TAMP system. We use the learned components for constraints and samplers that are difficult to engineer in the planning model, and use a TAMP solver to search for the task plan with constraint-satisfying action parameter values. To tractably make predictions for unseen objects in the environment, we define the learned samplers and TAMP operators on learned latent embedding of changing object states. We evaluate our approach in a simulated articulated object manipulation domain and show how the combination of classical TAMP, generative modeling, and latent embedding enables multi-step constraint-based reasoning. We also apply the learned sampler in the real world. Website: https://sites.google.com/view/dimsam-tamp.

Effective execution of long-horizon tasks with dexterous robotic hands remains a significant challenge in real-world problems. While learning from human demonstrations has shown encouraging results, they require extensive data collection for training. Hence, decomposing long-horizon tasks into reusable primitive skills is a more efficient approach. To achieve so, we developed DexSkills, a novel supervised learning framework that addresses long-horizon dexterous manipulation tasks using primitive skills. DexSkills is trained to recognize and replicate a select set of skills using human demonstration data, which can then segment a demonstrated long-horizon dexterous manipulation task into a sequence of primitive skills to achieve one-shot execution by the robot directly. Significantly, DexSkills operates solely on proprioceptive and tactile data, i.e., haptic data. Our real-world robotic experiments show that DexSkills can accurately segment skills, thereby enabling autonomous robot execution of a diverse range of tasks.