Pioneering Minds Redefine the Future of AI, Robotics, and Space

From microscopic optical systems to self-assembling space stations, a new generation of researchers is building the foundational technologies that will power our AI-driven future.

Miniaturizing Light: The Dawn of Silicon Photonics

Jelena Notaros, a visionary researcher at MIT, is at the forefront of a technological revolution in optical engineering. Her work focuses on significantly shrinking complex optical systems, integrating them onto tiny silicon wafers. This groundbreaking field, known as silicon photonics, promises to transform how electronic devices are designed and manufactured. Traditionally, optics conjured images of bulky lenses and intricate fiber optic setups confined to laboratories. Notaros and her team are challenging this perception, demonstrating that tens of thousands of optical components can be condensed onto a single computer chip, no larger than a few millimeters.

The advantages of this miniaturization are multifaceted. By dramatically reducing the size of optical components, devices become more compact, consume less energy, and offer greater design flexibility. Furthermore, the manufacturing process leverages standard lithography techniques, the same methods used to produce conventional computer chips. This inherent scalability and cost-efficiency open pathways for widespread adoption across various industries. To illustrate the potential, Notaros and her collaborators successfully miniaturized a 3D printer to the approximate size of a quarter, subsequently using it to print the MIT logo at a mere one millimeter in height. Their innovative efforts also extended to creating a tiny optical tweezer, a type of tractor beam, which they employed to precisely manipulate the shape of a mouse cell. In more recent endeavors, Notaros's team developed a chip that contained and cooled charged particles close to absolute zero, using them to encode information for quantum computing. On a more practical, human-centric level, they engineered a prototype for augmented reality glasses capable of projecting holographic images that eliminate common issues like eyestrain and headaches, a project that has garnered support from the US military through the Defense Advanced Research Projects Agency.

However, one of the most immediate and impactful applications of Notaros’s work is anticipated in the realm of lidar sensors. These sensors, currently recognizable as the prominent, often spinning, units on autonomous vehicles like Waymo cars, are crucial for perceiving environments. Notaros envisions replacing their moving parts and shrinking the entire sensor down to the size of an uncooked lentil. This solid-state lidar design would not only be discreetly integrated into multiple points around a vehicle but also boast superior capabilities, including the detection of objects at significantly greater distances than current conventional designs.

Reversible Computing: A Quest for Energy Efficiency

The ubiquitous computer chips that power our digital world inherently waste a substantial portion of the energy they consume, dissipating it as heat. This energy loss not only limits processing speeds but also necessitates extensive cooling systems, ranging from the fans in our personal laptops to the colossal, water-intensive infrastructure of AI data centers. Hannah Earley, Chief Technology Officer and co-founder of Vaire Computing, is pioneering a transformative approach to computing that aims to address this fundamental inefficiency: reversible computing.

Conventional computing operates in a forward-only manner. Information is processed, energy is expended, and then data no longer needed is discarded, with a portion of the energy lost as heat. Reversible computing fundamentally alters this paradigm. Instead of simply discarding information, it seeks to reverse calculations, effectively retaining and reusing energy that would otherwise be wasted. This concept is akin to a closed-loop system, where energy is recirculated rather than being allowed to escape. In 2025, Earley and her team unveiled a groundbreaking proof-of-concept circuit that demonstrated the viability of this approach for real computations in hardware. Crucially, their tests showed net energy recovery, indicating that the energy saved at least matched the energy required for the recovery process. This achievement unequivocally proved that computations do not have to inherently result in significant electrical energy loss as heat.

The path ahead for Vaire Computing involves demonstrating that these energy savings can be sustained in more complex, practical computing environments and that their technology can seamlessly integrate with existing chipmaking infrastructure. Despite these significant challenges, Earley remains optimistic. She asserts that the underlying physics of reversible computing points to it as the optimal method for constructing future computers. Should they succeed, reversible computing holds the profound potential to decouple the relentless growth of computing power from an ever-increasing demand for electricity, offering a sustainable path forward for the digital age.

Robots That See and Learn: A Leap Towards General Intelligence

Modern robots, despite their impressive capabilities, often operate with a limited form of intelligence. They excel at repetitive, precisely defined tasks within controlled environments, akin to a specialized kitchen appliance. However, introduce a new setting, alter the tools, or modify the ingredients, and their performance can quickly falter. Shuang Li, a research scientist at Google DeepMind, is dedicated to endowing robots with a more adaptable and flexible form of intelligence.

During her doctoral research at MIT, Li made significant contributions to early studies demonstrating how language models could guide robotic decision-making. In a static environment, an AI agent could articulate the correct sequence of actions to achieve a specific goal. Yet, planning is only one facet of the challenge. A robot tasked with placing an apple in a refrigerator, even if it knows the steps, must first identify both the apple and the refrigerator, and then execute a series of actions in an environment that may visually differ significantly from its training examples. To address this, during her postdoctoral tenure at Stanford, Li spearheaded the development of Unified Video Action (UVA). This innovative training model empowers robots to transfer learned behaviors between diverse settings. UVA training videos instruct robots not only on the actions to perform but also on how those actions will alter the robot's visual perception. For instance, when picking up a cup, the robot learns to concentrate intently on the cup, tracking its movement upward, rather than becoming distracted by background elements or even its own manipulating hand. This focused attention helps prevent the robot from becoming disoriented by unfamiliar surroundings.

Released in 2025 as an open-source package, UVA now underpins Li's work at Google DeepMind. Her current focus is on integrating the sophisticated planning capabilities of language models with the intuitive physical understanding provided by video models into a unified system. Li harbors strong hopes that as robots are engineered for increasingly intricate tasks and the costs associated with training them decrease, models like UVA will be instrumental in transforming robots into highly capable assistants, whether in a domestic kitchen or a complex industrial environment. Jianlan Luo, from the Shanghai Innovation Institute, shares this ambition, specifically focusing on using AI to enable robots to learn new tasks more rapidly and efficiently. His work explores methods like reinforcement learning, where robots are trained in simulations or from task demonstration videos, and then practice independently. This continuous learning, guided by AI, accelerates skill acquisition and adaptability, further enhancing the potential for robots to integrate seamlessly into diverse operational settings.

Building Beyond Earth: Habitats in the Cosmos

The construction of extraterrestrial infrastructure, such as space stations, has largely remained static for decades. The size of each component is constrained by a rocket's cargo capacity, and astronauts, clad in cumbersome spacesuits, meticulously assemble these pieces while tethered. This process is inherently slow, perilous, and prone to error. Ariel Ekblaw, founder of the Aurelia Institute, is pioneering a revolutionary approach to space construction that replaces human manual labor with autonomous, self-assembling parts, promising cheaper, more scalable designs and significantly safer construction methodologies.

Ekblaw, who completed her PhD in robotics at MIT, designed hexagonal and pentagonal tiles intended to serve as fundamental building blocks. These tiles can be loaded in bulk into a spacecraft. Once released into the vacuum of space, the tiles, embedded with rows of magnets along their edges, autonomously connect in a predetermined fashion. These modular components offer immense versatility, capable of configuring into a wide array of structures, from walkways and living quarters to, potentially in the future, orbital solar collectors and data centers. To bring this technology to fruition, Ekblaw co-founded two entities: the non-profit Aurelia Institute, dedicated to constructing human habitats in space, and the for-profit Rendezvous Robotics, focused on leveraging these tiles for space-based solar collectors and data centers.

The self-assembly process was successfully demonstrated in 2020 and again in 2022 aboard the International Space Station (ISS) using small-scale models. The next critical step is anticipated later this year, when 32 dinner-plate-sized tiles will self-assemble into a spherical structure inside the ISS. If this demonstration proves successful, a larger set of tiles is slated for an in-space demonstration outside the ISS in 2027. Ekblaw hopes that within a year or two following that, Rendezvous Robotics will launch early designs into orbit for commercial testing. Her long-term vision is for these tiles to form an add-on module to whatever platform succeeds the ISS, which is scheduled for decommissioning in 2030, marking a new era of agile and scalable space infrastructure.

Why it matters

The innovations highlighted in this briefing underscore a profound shift in technological development, with direct and transformative implications for in-field AI, technicians, telco, and data center operations. Silicon photonics, as advanced by Jelena Notaros, promises to shrink the footprint and energy demands of optical components, enabling more powerful and compact sensors for robotic systems and potentially integrating high-speed optical networking directly into smaller devices. This could lead to a new generation of edge AI hardware for remote field technicians, offering advanced diagnostic and augmented reality tools without the bulk or power drain of current systems. For telco infrastructure, these miniaturized, efficient optical systems could revolutionize data transmission within smaller, distributed network nodes, reducing the need for extensive cooling. Hannah Earley's work on reversible computing directly addresses the escalating energy consumption of data centers, a critical concern for both financial viability and environmental impact. By drastically reducing heat generation, reversible computing could enable denser, more powerful AI training clusters and compute servers, alleviating the massive cooling infrastructure currently required. This directly impacts data center operations by lowering operational costs and increasing computational capacity within existing physical footprints. Shuang Li's and Jianlan Luo's advancements in robotic intelligence, particularly in areas like adaptable learning and perception, are crucial for deploying more versatile robots in complex environments, such as maintaining remote telco towers or assisting in data center hardware installation and troubleshooting. These robots could perform tasks with greater autonomy and less human intervention, enhancing safety and efficiency for field technicians. Finally, Ariel Ekblaw's self-assembling space habitats, while seemingly distant, hint at future orbital data centers and communication platforms. The technologies for self-assembly, modularity, and autonomous construction could translate into terrestrial applications for rapid deployment of infrastructure, potentially benefiting disaster recovery efforts or the quick setup of temporary telco networks and mobile data centers. These pioneers are not just inventing new gadgets; they are laying the groundwork for a more efficient, intelligent, and interconnected world, impacting every facet of our digital and physical infrastructure.