International Journal of Humanoid Robotics

In this editorial dedicated to the special issue which celebrates the 20th anniversary of the International Journal of Humanoid Robotics (IJHR), we outline the history of IJHR.

A humanoid robot is a particular form of embodied agent. The form that an agent takes has a major impact on how that agent interacts with its environment and how it develops an understanding of that environment through its interactions. In this paper, we explore the importance of humanoid embodiment and we argue that humanoids occupy a special niche in the spectrum of robot forms. In doing so, we highlight the implications for the way a humanoid robot can interact with its environment, including humans, for the manner in which humans interact with humanoid robots, and for a humanoid robot’s capacity to develop cognitive abilities. We also consider the degree to which humanoid robots should approximate humans, addressing robot morphology, appearance, and movement. We emphasize the dual role of humanoid robots as engineering artifacts that can provide services for humans, and as platforms for scientific enquiry into the nature of human cognition. We conclude by highlighting some key research challenges for the discipline of humanoid robotics.

The most fundamental oversight in Artificial Intelligence (AI) is probably the avoidance of conscious learning. The most widespread misconduct in AI seems to be Post-Selection. Avoiding conscious learning, we program humanoid robots that do not have the intent to learn new skills, such as standing up, walking, jumping, speaking, and thinking. Because programmed-in skills are all brittle, we must imitate natural intelligence, from insects to humans, that all learn from consciousness. Worldwide AI researchers and the public have been misled by media hypes about false AI performances, from Deep Learning to ChatGPT, rooted in Post-Selection. However, the Post-Selection protocol behind such hypes is fatally flawed by alleged misconduct — Misconduct 1, Cheating in the absence of a test; Misconduct 2, hiding bad-looking data. In other words, the reported errors are only data-fitting errors, instead of testing errors. This paper discusses how Conscious Learning is enabled by Developmental Network 3 (DN-3) to learn from its own intents without Post-Selection misconduct. Among many new concepts presented here, this paper establishes a new theorem that the expected error of the luckiest system in a future test is the same as any other less lucky system, namely, average. In contrast, DN-3 develops a sole network that is optimal in the sense of maximum likelihood (ML), better than the luckiest system on a validation set. The ML optimality transfers the performance on a validation set in the prior lifetime to a test set in the future lifetime. Many other AI techniques, e.g., symbolic, connectionist, and evolutional, also use Post-Selection which lacks such a transfer.

The purpose of this paper is to present my personal journey in the last 30-plus years of research and development efforts on a cognitive robotic system called Intelligent Soft Arm Control (ISAC). In doing so, I will highlight how our team approached solving certain robot intelligence, communication, and control problems.

This paper explores the confluence of physical embodiment and social interaction in the context of Cognitive Developmental Humanoid Robotics (CDHR). By classifying varied interactions through developmental stages of the “self” and their interaction spheres, the discussion unearths profound insights into the composite nature of developmental processes. It presents a multi-dimensional exploration through different interaction scenarios, ranging from fetus-mother interactions to advanced large language models like ChatGPT, revealing the intrinsic connection between the physical and social realms of existence. In conclusion, this paper broadens the horizon of our understanding of the intricate interplays between physical embodiment and social interaction, setting the stage for more nuanced, ethically sound approaches and explorations in the realm of humanoid robotics and artificial intelligence.

Embodiment is a key to solving the reliability and alignment issues in the current AI. This is because it imposes consistent constraints on the entire agent-environment interactions and accompanying information without specifying their actual contents, and the constraints are common to those with similar embodiment. The concept of embodiment should be generalized beyond mechanical properties of the body and information structure of sensory signals, encompassing internal organs and metabolism, mental processes, and inter-agent interactions, to “super-embodiment”. It can address sensibilities, values and morals toward artificial humanity, which will be critically important for the next generation AI. [Y. Kuniyoshi, Open-ended intelligence arises from emergence and development from embodiment, J. Robotics Soc. Jpn. 41(7) (2023) 585–590 (in Japanese); Y. Kuniyoshi, Fusing autonomy and sociality via embodied emergence and development of behaviour and cognition from fetal period, Philos. Trans. R. Soc. B: Biol. Sci. 374 (2019) 20180031].

The interplay between computational neuroscience and humanoid robotics, both historically and looking forward, is presented, with a particular focus on two significant projects: the ERATO Kawato Dynamic Brain and the ICORP Computational Brain. Discussion revolves around pivotal brain regions, namely the cerebellum, basal ganglia, and cerebrum, and the associated computational models.

In 1997, Sony announced AIBO, a fully autonomous small quadruped robot for home entertainment, and in 1999 the company began selling it as a consumer product. Soon after development, two small humanoid robots were announced. One was QRIO by Sony, which is about 60cm height body with dynamical bipedal walking and whole-body cooperative control. The other was PINO by the ERATO Kitano Symbiotic Systems Project, which is about 70cm height body, Open HW/SW with inexpensive off-the-shelf component. In this paper, we revisit the two humanoids and nearly 20-year-old technologies, and discuss what were done 20 years ago, what have been achieved and what challenges are ahead.

Twenty years ago in the first issue of IJHR, among others, there was a published paper [M. Vukobratović and B. Borovac, Zero-moment point — Thirty five years of its life, Int. J. Humanoid Robot. 1(1) (2004) 157–173], devoted to well-known zero-moment point (ZMP) concept. Its role in biped locomotion (in the preservation of dynamic balance during walking and gait synthesis) was (and still is) extremely significant. This paper is aimed to present, how far we have gone in the past 20 years in the realization of bipedal walking, what are remaining open challenges and what we can expect in the further development. At the beginning of the paper, we will briefly refer to some still existing misinterpretations of the ZMP notion, related to specific boundary cases, when the terms Center of Pressure (CoP) and ZMP are still being mixed. Then, we will point your attention to the discussion of two important issues: attempts of generalization of ZMP notion in case of motion in the presence of multiple, spatially distributed and non-planar contacts and walk realization in a non-laboratory environment (unstructured environment and on the non-flat ground surface). Of course, looking at “how humans do it” has always been an inspiration for the realization of complex tasks by humanoids. Learning to walk is a task where several successful realizations have appeared only recently, so we will look at this topic at the end of this paper.

Teachability has been extensively studied under the context of making industrial robots to be programmable and reprogrammable. However, it is only recently that the artificial intelligence (AI) research community is accelerating the research works with the objective of making humanoid robots and many other robots to be teachable under the context of using natural languages. We human beings spend many years learning knowledge and skills despite our extraordinary mental capabilities of being teachable with the use of natural languages. Therefore, if we would like to develop human-like robots such as humanoid robots, it is inevitable for us to face the issue of making future humanoid robots teachable with the use of natural languages as well. In this paper, we present the key details of a top-down design for achieving a teachable mind which consists of two major processes: the first one is the process that enables humanoid robots to gain innate mental capabilities of transforming incoming signals into meaningful crisp data, and the second one is the process which enables humanoid robots to gain innate mental capabilities of undertaking incremental and deep learning with the main focus of associating conceptual labels in a natural language to meaningful crisp data. These two processes consist of the two necessary and sufficient conditions for future humanoid robots to be teachable with the use of natural languages. In addition, this paper outlines a very likely new finding underlying the human brain’s neural systems as well as the obvious mathematics underlying artificial deep neural networks. These outlines provide us with a strong reason to separate the study of the mind from the study of the brain. Hopefully, the content discussed in this paper will help the AI research community to venture into the right direction which is to make future humanoid robots, non-humanoid robots, and many other systems to achieve human-like self-intelligence at the cognitive level with the use of natural languages.