Figure 1 - uploaded by Satoshi Kagami
Content may be subject to copyright.
Source publication
This paper gives an overview of the humanoid robot 'H7', which was developed over several years as an experimental platform for walking, autonomous behaviour and human interaction research at the University of Tokyo. H7 was designed to be a human-sized robot capable of operating autonomously in indoor environments designed for humans. The hardware...
Contexts in source publication
Context 1
... standard dynamic programming techniques, optimal sequences of footstep placements can be computed according to encoded heuristics that minimize the number and complexity of the steps taken. Such a strategy can be computed efficiently on standard PC hardware (under 1 s for simple environments and in a few seconds for relatively complex, cluttered environments, as shown in figure 11). ...
Context 2
... biped model comes with a pre-determined set of feasible footstep locations for each foot. For example, figure 12 shows the continuous feasible footstep range FR right for the right foot while supported by the left foot, and an example discrete set of foot placements. For symmetric bipeds, the placements for the left foot can simply mirror the right-foot placements. ...
Context 3
... of the terms is weighted relative to each other by the factors w d , w r and w g . Figure 11 shows a cluttered office in which a model of the humanoid robot must navigate and a top view of a footstep sequence computed to reach a circular goal region in the centre of the room. There were a total of 15 discrete foot placements considered for each foot and a total of 20 floor obstacles. ...
Context 4
... path was computed in approximately 4 s on a 1.6 GHz Pentium4 running LINUX. We used a two-dimensional polygon- polygon intersection test for the first phase of collision checking, and the (V-clip) library (see Mirtich 1998) for fast minimum distance determination between the obstacles and the convex hull of each leg link for the second phase ( figure 13). ...
Context 5
... a graphical user interface, an operator can click and drag an object to a target location and issue a move command. Figure 14 shows snapshots of a planned motion for a humanoid repositioning a bottle from the lower shelf to the upper shelf. In the examples shown in figure 15, the simulated vision module is used in order to verify that a particular target object is visible to a virtual humanoid prior to attempting to grasp it. ...
Context 6
... 14 shows snapshots of a planned motion for a humanoid repositioning a bottle from the lower shelf to the upper shelf. In the examples shown in figure 15, the simulated vision module is used in order to verify that a particular target object is visible to a virtual humanoid prior to attempting to grasp it. If the object is visible, the manipulation planner is invoked to plan a collision-free path to grasp the object. ...
Context 7
... have tested the output trajectories calculated by the planner online. Figure 16 shows a computed dynamically stable motion for the robot moving from a neutral standing position to a low crouching position in order to retrieve an object from beneath a chair. Figure 17 shows a motion for positioning the right leg above the top of a box while balancing on the left leg. ...
Context 8
... 16 shows a computed dynamically stable motion for the robot moving from a neutral standing position to a low crouching position in order to retrieve an object from beneath a chair. Figure 17 shows a motion for positioning the right leg above the top of a box while balancing on the left leg. Each of the scenes contains over 9000 triangle primitives. ...
Context 9
... a reference region (region A) is selected from the left image, and the right image is searched for the best matching region (region B), search area reference area matching left image r i ght image Figure 18. Online consistency checking method. ...
Context 10
... the correlation value is calculated locally, the first two steps of the consistency check can be calculated simultaneously, followed by the third step that calculates the best match. Figure 18b illustrates the process. ...
Context 11
... desired position to which the robot should navigate towards is converted to a forward distance a and lateral distance b. A continuum of candidate positions for the robot destination point to reach a fixed distance from the target is shown in figure 19. One destination point is chosen from among the candidates by calculating the point on the line connecting the target and the torso position at the end of the currently executing step ( P c in figure 19). ...
Context 12
... continuum of candidate positions for the robot destination point to reach a fixed distance from the target is shown in figure 19. One destination point is chosen from among the candidates by calculating the point on the line connecting the target and the torso position at the end of the currently executing step ( P c in figure 19). The desired torso translation for the next single step is given as P d K P c . ...
Citations
... Similarly, the Atlas humanoid robot developed by Boston Dynamics failed while doing parkour (Pescovitz, 2021). Many other humanoids that closely approximate neuro-robots include: the H-7 of the University of Tokyo (Nishiwaki et al., 2007), Asimo of Honda Motor Corporation (ASIMO, 2011), Qrio of Sony Corporation (Carnegie Mellon Today, 2005), DB from Utah-based company Sarcos (Cheng, 2015), or HRP-2 from the Japanese Humanoid Robotics Program (Hirukawa et al., 2004), among others. Although representing high technological achievements, these robots have yet to IHMC's Atlas was one of the robots that fell during the DARPA robotics challenge finals. ...
Neuro-robots are a class of autonomous machines that, in their architecture, mimic aspects of the human brain and cognition. As such, they represent unique artifacts created by humans based on human understanding of healthy human brains. European Union’s Convention on Roboethics 2025 states that the design of all robots (including neuro-robots) must include provisions for the complete traceability of the robots’ actions, analogous to an aircraft’s flight data recorder. At the same time, one can anticipate rising instances of neuro-robotic failure, as they operate on imperfect data in real environments, and the underlying AI behind such neuro-robots has yet to achieve explainability. This paper reviews the trajectory of the technology used in neuro-robots and accompanying failures. The failures demand an explanation. While drawing on existing explainable AI research, we argue explainability in AI limits the same in neuro-robots. In order to make robots more explainable, we suggest potential pathways for future research.
... To improve the rolling characteristics and increase the step length, some robots include a toe joint with passive stiffness in the flat foot design. For example, the foot of the humanoid M-Series [32], H6 [33], H7 [34], COMAN [35], and cCub [36] are designed for these characteristics. The humanoid robot LOLA [37] comprises a similar design with an active toe joint that allows the leg swing to be in a more extended configuration. ...
Robotic exploration in natural environments requires adaptable, resilient, and stable interactions with uncertain terrains. Most state-of-the-art legged robots utilize flat or ball feet that lack adaptability and are prone to slip due to point contact with the ground. In this article, we present guidelines to design an adaptive foot that can interact with the terrain to achieve a stable configuration. The foot is inspired by goat hoof anatomy that incorporates roll and yaw rotations in the Fetlock and Pastern joints, respectively. To ensure adaptability with stability in physical interaction and to prevent the foot from collapsing, we provide a lower bound on each joint's stiffness. In addition, we also render an upper bound to conform to the high force exchange during interactions with the ground consisting of certain roughness. Based on these guidelines, we design the hoof and experimentally validate the theoretical results with a loading test setup in lab settings. We use four different friction materials with various triangular, rectangular, and semicircular extrusions to simulate common ground features. We observe that hooved pads require more load for the system to be unstable. Any anatomically inspired foot can be designed based on the guidelines proved analytically and experimentally in this article.
... Huang et al. [13] design a crawling strategy for a biped walking with a rigid-flexible waist using CPG control. Earlier works involve controlling humanoid robots to crawl with hands and feet [14], [15] or hands and knees [16], [17], [18], [19]. However, these robots are either mechanically complicated or energy-consuming. ...
... III-E on the robot. The values of e threshold in (14,15) in our experiment are 0.15 and 0.35 for e pitch and The results show that ARRL outperforms pure RL methods (Vanilla TD3 and Vanilla SAC) on real-world implementation, especially after applying the progressive sim-to-real strategy: four ARRL combinations reach a walking distance greater than 1 meter, while the majority has a walking distance better than pure RL methods. Although ARRL is not the best after the sim-to-real transfer-pure black-box optimisers (Vanilla TBPSA, BO, and CMAES) were capable of reaching the highest scores-it still shows a very good sim-to-real transferability. ...
While quadruped robots usually have good stability and load capacity, bipedal robots offer a higher level of flexibility / adaptability to different tasks and environments. A multi-modal legged robot can take the best of both worlds. In this paper, we propose a multi-modal locomotion framework that is composed of a hand-crafted transition motion and a learning-based bipedal controller -- learnt by a novel algorithm called Automated Residual Reinforcement Learning. This framework aims to endow arbitrary quadruped robots with the ability to walk bipedally. In particular, we 1) design an additional supporting structure for a quadruped robot and a sequential multi-modal transition strategy; 2) propose a novel class of Reinforcement Learning algorithms for bipedal control and evaluate their performances in both simulation and the real world. Experimental results show that our proposed algorithms have the best performance in simulation and maintain a good performance in a real-world robot. Overall, our multi-modal robot could successfully switch between biped and quadruped, and walk in both modes. Experiment videos and code are available at https://chenaah.github.io/multimodal/.
... 18 Several algorithms that basically lie in the domains of environment perception, robot localization, and environment mapping have been explored to develop some autonomous behaviors for humanoid robots. 19,20 Also, probabilistic approaches helped the robot to construct maps for different environments and localize itself in the environment using simultaneous localization and mapping algorithms and sensor fusion paradigms like extended kalman filter (EKF) and particle filter-based approaches. 21 One of the most challenging aspects of humanoid robot design is the technique used in walking pattern generation since a biped robot is an inherently unstable structure in essence. ...
This article presents the modeling process of the lower part of a humanoid biped robot in terms of kinematic/dynamic states and the creation of a full dynamic simulation environment for a walking robot using MATLAB/Simulink. This article presents two different approaches for offline walking pattern generation: one relying on a closed-form solution of the linear inverted pendulum model (LIPM) mathematical model and another that considers numerical optimization as means of desired output trajectory following for a cart table state-space model. This article then investigates the possibility of introducing solution-dependent modifications to both approaches that could increase the reliability of basic walking pattern generation models in terms of smooth single support–double support phase transitioning and power consumption optimization. The algorithms were coded into offline walking pattern generators for NAO humanoid robot as a valid example and the two approaches were compared against each other in terms of stability, power consumption, and computational effort as well as against their basic unmodified counterparts.
... An early-contact concept is especially important for robots with position-controlled joints with the resulting high, intrinsic stiffness. There are several ways to mitigate the impact effects of the swing foot on uneven terrain, such as the modification of the joint controller gains based on the gait cycle timing 24,25 or the activation of a ground reaction force controller once the swing foot is in contact. 10 Another strategy is the implementation of reflexive heuristics that change the vertical trajectory of the swing foot when an early contact occurs-that is, when the vertical contact forces exceed a certain threshold. ...
... The method described with (24) can already effectively accelerate the CoM to generate missing contact forces. The approach is, however, local in time and not limited to certain walking phases. ...
... One approach for robots with position-controlled joints is to reduce the swing leg's joint controller gains just before anticipated contact. 24,25 Alternatively, admittance control can be implemented by modifying the desired trajectories based on FT sensor information. 18,19,[32][33][34] The force reference for this controller usually depends on a predefined 18,32,34 or previously recorded force profile. ...
The robustness of biped walking in unknown and uneven terrains is still a major challenge in research. Traversing such environments is usually solved through vision-based reasoning on footholds and feedback loops—such as ground force control. Uncertain terrains are still traversed slowly to keep inaccuracies in the perceived environment model low. In this article, we present a ground force-control scheme that allows for fast traversal of uneven terrain—including unplanned partial footholds—without using vision-based data. The approach is composed of an early-contact method, direct force control with an adaptive contact model, and a strategy to adapt the center of mass height based on contact force data. The proposed method enables the humanoid robot Lola to walk over a complex uneven terrain with 6 cm variation in ground height at a walking speed of 0.5 m/s. We consider our work a general improvement on the robustness to terrain uncertainties caused by inaccurate or even lacking information on the environment.
... Experiments and interaction studies are equipment intensive, time-consuming and labor intensive [2]. Availability of research platform, high cost and proprietary costs are additional challenges faced by academic researchers [3]. Furthermore, HMI experiments rarely run correct the first time [4], and although preparing all experimental parameters well in advance is ideal, this is not the case in practice. ...
... For novel interaction techniques in early stages of product development constructing a well-defined hypotheses if often very hard [5]. For example, multiple years were required to develop a humanoid research platform investigating human interaction [3]. Experimental setups like the car simulation of Ahn et al. [6] and autonomous car simulation of Gil et al. [7] are likely to be quicker to set up and pilot. ...
This work presents a modular research platform to design, test and run human-machine interaction (HMI) experiments. Traditionally, HMI experiments are time and resource consuming, particularly in the piloting phase. Furthermore, such experiment setups are often rigid and only fit to one particular hypothesis. Thus, significant time is needed to alter the setup to new hypotheses, if this is possible at all. The platform presented is a technical proof-of-concept of a highly flexible experiment setup, which can rapidly be adapted to alternative hypotheses. Examples of interchangeable modules include simulator software (context), user interface (independent variable) and human operator physiology sensors (dependent variable). An agile product development methodology, Wayfairing, was used to accomplish this.
... Providing services such as disabled person carrying, housework assistance are classical examples for smooth interactions. Moreover, some projects aim more complex tasks such as ultimate helpers in man-made or natural disasters [1][2][3][4]. In this context, safe robot motion control is the point that slows down the design of the humanoid robot, and the fulfillment of these kinds of tasks. ...
In this paper, we present a new type of low cost force localized interaction sensor dedicated to robot interactive sensing system. The design of the sensor is based on mechanical deformation and mathematical models. It includes the capability to be integrated easily in robotic limbs. This sensing system is based on the coupled results obtained from six well-located force sensors in order to measure the magnitude, position and orientation of the interacting force. The mathematical model to calculate the force from the measurements is fully detailed. A prototype in the 2-Dimensional plan is built. An experimental proof of concept setup is illustrated and sets of measurements illustrate force sensing results. A commercial sensor based on another technology is used for reference. The comparison with the reference shows very close results on a tested range up to 50 N. The standard deviation of the difference between both sensors is lower than 1 N on a tested range of 50 N and for several angles of the applied force, demonstrating the relevance of the new sensor.
... However, they must be robust as they are subject to repeated impacts. Many humanoid robots incorporate F/T sensors at the ankles to measure forces and moments associated with maintaining balance and sense ground reaction forces (GRFs) [12]- [16]. For the running Cheetah robot, Chuah et al. [17] present a light and robust foot sensor. ...
... Individual normal taxel (N1-5) force peak amplitude [13][14][15][16][17] Individual normal taxel (N1-5) rise time to peak [18][19][20][21][22] Ratio of individual taxel force peak to peak sum(N1-5) 23-25 ...
For small legged robots, ground contact interactions significantly affect the dynamics and locomotion performance. In this article, we designed thin, robust capacitive tactile sensors and applied them to the feet of a small hexapod with C-shaped rotating legs. The sensors measure contact forces as the robot traverses different types of terrain including hard surfaces with high or low friction, sand, and grass. Different gaits perform best on different types of terrain. Useful measured parameters include the magnitude and timing of the peak normal forces, in combination with the leg rotational velocity. The measured parameters were used in a support vector machine classifier to identify terrain types with 82.5% accuracy. Based on gait performance studies, we implemented a terrain-based gait control using real-time terrain classifications. A surface transitioning test shows 17.1% increase in body speed and 13.2% improvement in efficiency as the robot adjusts its gait.
... An early-contact concept is especially important for robots with position-controlled joints with the resulting high, intrinsic stiffness. There are several ways to mitigate the impact effects of the swing foot on uneven terrain, such as the modification of the joint controller gains based on the gait cycle timing 24,25 or the activation of a ground reaction force controller once the swing foot is in contact 10 . Another strategy is the implementation of reflexive heuristics that change the vertical trajectory of the swing foot when an early contact occurs -i.e. when the vertical contact forces exceed a certain threshold 8,26 . ...
... The method described with (24) can already effectively accelerate the CoM to generate missing contact forces. The approach is, however, local in time and not limited to certain walking phases. ...
... Consequently, the control law (24) changes to ...
The robustness of biped walking in unknown and uneven terrains is still a major challenge in research. Traversing such environments is usually solved through vision-based reasoning on footholds and feedback loops-such as ground force control. Uncertain terrains are still traversed slowly to keep inaccuracies in the perceived environment model low. In this article, we present a ground-force control scheme that allows for fast traversal of uneven terrain-including unplanned partial footholds-without using vision-based data. The approach is composed of an early-contact method, direct force control with an adaptive contact model, and a strategy to adapt the center of mass height based on contact force data. The proposed method enables the humanoid robot LOLA to walk over a complex uneven terrain with 6 cm variation in ground height at a walking speed of 0.5 m/s. We consider our work a general improvement on the robustness to terrain uncertainties caused by inaccurate or even lacking information on the environment.
... A quasi or semi-passive walker based on the passive walker (McGeer 1993) was designed to utilize its inertial force and weight by carefully shaping its round foot and setting the center of mass (CoM). Therefore, such a robot equips a minimal number of actuators to bend a knee and swing a leg to realize a successful locomotion even on a flat plane (Wisse and Van Frankenhuyzen 2002;Collins et al. 2001;Anderson et al. 2004;Dertien 2006) and is more energy efficient (Collins et al. 2005) than bipedal robots whose design is not based on the passive walker (Akachi et al. 2005;Park et al. 2004;Nishiwaki et al. 2007). However, the walking behavior, such as the walking velocity or cycle, of passive and semipassive walkers cannot be controlled because it depends on a fixed physical property, such as the CoM and the foot shape, B Takashi Takuma takashi.takuma@oit.ac.jp 1 Osaka Institute of Technology, 5-16-1, Omiya, Asahi-ku, Osaka, Japan while the current bipedal walkers driven by electric servo motor control their walking velocity by changing trajectories of joints. ...
A semi-passive walker that equips a small number of actuators attains locomotion through an interaction between its physical properties such as center of mass (CoM) and the ground in contact. Therefore, it is expected that the robot obtains a wide range of locomotion such as in its walking velocity by changing a trajectory of CoM. In this study, we focused on the trunk mechanism that largely influences the walking behavior by changing its CoM, and propose a vertebrae-inspired trunk mechanism in which tunable viscoelastic joints are embedded. By switching a mechanical property, that is, the viscoelasticity of the trunk joints, the cycle of the passive oscillation while walking is changed, and the robot obtains a wide range of the walking cycle. To verify this trunk mechanism, we developed a physical bipedal robot. The physical trunk mechanism does not require a supplemental actuator to oscillate the trunk actively, and does not require energy consumption to retain the viscoelasticity according to the devised mechanism. We also constructed simulation models equipping various types of trunk mechanisms for verifying the number and position of the trunk joints that influence the variation of the walking cycle. The simulation results suggest a criterion for the trunk mechanism design in which the walking cycle is influenced by the number and position of the joints. This paper concludes that the proposed trunk mechanism is a suitable and practical mechanical element that provides a wide range of the locomotion of semi-passive walker by utilizing its switchable mechanical property.
