Vision-based reinforcement learning (RL) is a generalizable way to control an agent because it is agnostic of specific hardware configurations. As visual observations are highly entangled, attempts for vision-based RL rely on scene representation that discerns individual entities and establishes intuitive physics to constitute the world model. However, most existing works on scene representation learning cannot successfully be deployed to train an RL agent, as they are often highly unstable and fail to sustain for a long enough temporal horizon. We propose ASIMO, a fully unsupervised scene decomposition to perform interaction-rich tasks with a vision-based RL agent. ASIMO decomposes agent-object interaction videos of episodic-length into the agent, objects, and background, predicting their long-term interactions. Further, we explicitly model possible occlusion in the image observations and stably track individual objects. Then, we can correctly deduce the updated positions of individual entities in response to the agent action, only from partial visual observation. Based on the stable entity-wise decomposition and temporal prediction, we formulate a hierarchical framework to train the RL agent that focuses on the context around the object of interest. We demonstrate that our formulation for scene representation can be universally deployed to train different configurations of agents and accomplish several tasks that involve pushing, arranging, and placing multiple rigid objects.

Behavior-predefined adaptive control for heterogeneous continuum robots

The International Journal of Robotics Research, Vol. 44 , 1 : pp. 65 -95. Published January , 2025

Show Abstract

hide Abstract

Abstract

Continuum robots have great application value and broad prospects in various fields due to their dexterity and compliance. To fully exploit their advantages, it is crucial to develop an effective, accurate and robust control system for them. However, research on continuum robot control is still in its infancy and there are many problems remaining unsolved in this field. In particular, this paper focuses on the task-space behavior and the generic control of heterogeneous continuum robots. First, a controller is proposed to achieve the kinematic motion control and visual servoing of continuum robots with predefined task-space behavior. The predefined behavior is twofold: prescribed task-space error and predefined convergence time. Then, the proposed controller is integrated with a velocity-level kinematic mapping estimator to obtain a model-free control system, which is applicable to heterogeneous continuum robots. Furthermore, a re-adjustable performance function is proposed to ensure the effectiveness and robustness of the proposed control system in the presence of external disturbance. Finally, extensive simulations and experiments are performed based on heterogeneous continuum robots, including the cable-driven continuum robot, the parallel continuum robot, the concentric-tube robot, the flexible endoscope, and the pneumatic continuum robot. Our results demonstrate that the task-space error of heterogeneous continuum robots complies with the prescribed boundaries and converges to steady state in predefined time, which reveals the efficacy of the proposed control method.

MOB-Net: Limb-modularized uncertainty torque learning of humanoids for sensorless external torque estimation

The International Journal of Robotics Research, Vol. 44 , 1 : pp. 96 -128. Published January , 2025

Show Abstract

hide Abstract

Abstract

Momentum observer (MOB) can estimate external joint torque without requiring additional sensors, such as force/torque or joint torque sensors. However, the estimation performance of MOB deteriorates due to the model uncertainty which encompasses the modeling errors and the joint friction. Moreover, the estimation error is significant when MOB is applied to high-dimensional floating-base humanoids, which prevents the estimated external joint torque from being used for force control or collision detection in the real humanoid robot. In this paper, the pure external joint torque estimation method named MOB-Net, is proposed for humanoids. MOB-Net learns the model uncertainty torque and calibrates the estimated signal of MOB, substantially reducing the estimation errors of MOB. The external joint torque can be estimated in the generalized coordinate including whole-body and virtual joints of the floating-base robot with only internal sensors (an IMU on the pelvis and encoders in the joints). Furthermore, MOB-Net shows more robust performance for the unseen data compared to the end-to-end learning method, and the robustness of MOB-Net is validated through extensive simulations, real robot experiments, and ablation studies. Finally, various collision handling scenarios are presented to show the versatility of MOB-Net: contact wrench feedback control for locomotion, collision detection, and collision reaction for safety.

Reduced order modeling of hybrid soft-rigid robots using global, local, and state-dependent strain parameterization

The International Journal of Robotics Research, Vol. 44 , 1 : pp. 129 -154. Published January , 2025

Show Abstract

hide Abstract

Abstract

The need for fast and accurate analysis of soft robots calls for reduced order models (ROM). Among these, the relative reduction of strain-based ROMs follows the discretization of the strain to capture the configurations of the robot. Based on the geometrically exact variable strain parametrization of the Cosserat rod, we developed a ROM that necessitates a minimal number of degrees of freedom to represent the state of the robot: the Geometric Variable Strain (GVS) model. This model allows the static and dynamic analysis of open-, branched-, or closed-chain soft-rigid hybrid robots, all under the same mathematical framework. This paper presents for the first time the complete GVS modeling framework for a generic hybrid soft-rigid robot. Based on the Magnus expansion of the variable strain field, we developed an efficient recursive algorithm for computing the Lagrangian dynamics of the system. To discretize the soft link, we introduce state- and time-dependent basis, which is the most general form of strain basis. We classify the independent bases into global and local bases. We propose “FEM-like” local strain bases with nodal values as their generalized coordinates. Finally, using four real-world applications, we illustrate the potential of the model developed. We think that the soft robotics community will use the comprehensive framework presented in this work to analyze a wide range of specific robotic systems.

All Issues

Current Issue

OnlineFirst

The International Journal of Robotics Research

Table of Contents:

Data paper

YUTO MMS: A comprehensive SLAM dataset for urban mobile mapping with tilted LiDAR and panoramic camera integration

Abstract

Articles

ASIMO: Agent-centric scene representation in multi-object manipulation

Abstract

Behavior-predefined adaptive control for heterogeneous continuum robots

Abstract

MOB-Net: Limb-modularized uncertainty torque learning of humanoids for sensorless external torque estimation

Abstract

Reduced order modeling of hybrid soft-rigid robots using global, local, and state-dependent strain parameterization

Abstract

Contents

Sign in to my Account

Login with

Register for an Account

The International Journal of Robotics Research

Table of Contents:

Data paper

YUTO MMS: A comprehensive SLAM dataset for urban mobile mapping with tilted LiDAR and panoramic camera integration

Abstract

Articles

ASIMO: Agent-centric scene representation in multi-object manipulation

Abstract

Behavior-predefined adaptive control for heterogeneous continuum robots

Abstract

MOB-Net: Limb-modularized uncertainty torque learning of humanoids for sensorless external torque estimation

Abstract

Reduced order modeling of hybrid soft-rigid robots using global, local, and state-dependent strain parameterization

Abstract

Contents