Federico Pigozzi

Federico Pigozzi

Let us reflect on the state of robotics. This year marks the $101$-st anniversary of R.U.R., a play by the writer Karel Čapek, often credited with introducing the word "robot". The word used to refer to feudal forced labourers in Slavic languages. Indeed, it points to one key characteristic of robotic systems: they are mere slaves, have no rights, and execute our wills instruction by instruction, without asking anything in return. The relationship with us humans is commensalism; in biology, commensalism subsists between two symbiotic species when one species benefits from it (robots boost productivity for humans), while the other species neither benefits nor is harmed (can you really argue that robots benefit from simply functioning?). We then distinguish robots from "living machines", that is, machines infused with life. If living machines should ever become a reality, we would need to shift our relationship with them from commensalism to mutualism. The distinction is not subtle: we experience it every day with domesticated animals, that exchange serfdom for forage and protection. This is because life has evolved to resist any attempt at enslaving it; it is stubborn. In the path towards living machines, let us ask: what has been achieved by robotics in the last $100$ years? What is left to accomplish in the next $100$ years? For us, the answers boil down to three words: juice, need (or death), and embodiment, as we shall see in the following.

Federico Pigozzi

Biological agents possess bodies that are mostly of soft tissues. Researchers have resorted to soft bodies to investigate Artificial Life (ALife)-related questions; similarly, a new era of soft-bodied robots has just begun. Nevertheless, because of their infinite degrees of freedom, soft bodies pose unique challenges in terms of simulation, control, and optimization. Here we propose a novel soft-bodied agents formalism, namely Pressure-based Soft Agents (PSAs): they are bodies of gas enveloped by a chain of springs and masses, with pressure pushing on the masses from inside the body. Pressure endows the agents with structure, while springs and masses simulate softness and allow the agents to assume a large gamut of shapes. Actuation takes place by changing the length of springs or modulating global pressure. We optimize the controller of PSAs for a locomotion task on hilly terrain and an escape task from a cage; the latter is particularly suitable for soft-bodied agents, as it requires the agent to contort itself to squeeze through a small aperture. Our results suggest that PSAs are indeed effective at those tasks and that controlling pressure is fundamental for shape-changing. Looking forward, we envision PSAs to play a role in the modeling of soft-bodied agents, including soft robots and biological cells. Videos of evolved agents are available at https://pressuresoftagents.github.io.

Federico Pigozzi, Michael Levin

A hallmark of life on Earth is the ability of agents to exert causal power and be drivers of subsequent events. This is key to cognition at all scales. Causal emergence, measuring the degree to which an agent exerts unique predictive power on its future, is one consequence of causal power. Indeed, recent discoveries have shown that biological agents, even minimal ones, increase their causal emergence after learning new memories. However, there is a major knowledge gap regarding how causally emergent artificial agents are. We focused on Reinforcement Learning (RL) of neural-network agents across an array of environmental conditions, encompassing different algorithms, agent architectures, and six environments arranged on a complexity spectrum. For consistency, we computed the causal emergence of their latent-space representations over their lifetimes. We used the recently proposed ΦID to estimate causal emergence and tested how it related to learning performance. Our results suggested a Causally Emergent Alignment Hypothesis: successful agents exhibited causal emergence that was consistently predictive of final reward early in training and whose representational dynamics aligned with reward improvement in most tasks. This idea suggests that causal emergence may be a previously undisclosed axis of reorganization of neural representations in RL agents, with the potential to establish causal relationships and interventions that will lead to better RL agents. Our work also highlights the alignment between causal emergence and learning as another way biological and artificial creatures compare.