Tiny robots get smarter—and greener—with brainy new chips

MIT's Gleanmer chip represents a breakthrough in energy-efficient real-time 3D mapping by replacing traditional voxel cubes with adaptable Gaussian ellipsoids, a technique embodied in the GMMap algorithm. This innovative approach dramatically reduces memory and computational demands, enabling tiny, battery-powered robots to navigate complex environments using just about 6 milliwatts of power—roughly the energy consumption of a single LED. As Vivienne Sze, a lead researcher, explains, the chip's co-design of algorithm and hardware allows for storing large maps compactly and generating collision-free paths while using only 20% of the energy required by conventional methods.

Beyond robotics, the Gleanmer chip's ultra-low power profile and efficient mapping capabilities open new frontiers for lightweight augmented reality headsets and AI-driven applications. Its ability to operate continuously at just 6 milliwatts makes it ideal for extended use in medical simulation, industrial inspection, and detailed repair work, where prolonged battery life and precise spatial understanding are critical. This versatility underscores the chip's potential to revolutionize not only autonomous navigation but also wearable technology and complex blueprint interpretation.

By mid-2026, MIT had made significant strides in AI-driven control systems tailored for miniature flying robots ranging from pocket-sized group alpha vehicles to insect-scale group gamma drones. These tiny robots leverage millimeter-scale chips consuming mere hundreds of milliwatts, enabling precise localization and mapping through integrated cameras and inertial measurement units. This hardware-software synergy allows the drones to operate fully autonomously at speeds and complexities far beyond human pilot capabilities, managing tens of thousands of assets simultaneously and executing strategic maneuvers in real time.

In a breakthrough announced later in July 2026, MIT researchers unveiled a two-part AI control system that imparts insect-like agility to their miniature flying robots, boosting speed by approximately 450% and acceleration by 250%. This system combines a computationally efficient model-predictive controller that forecasts motion and selects optimal actions with a bioinspired framework, enabling the robots to perform complex maneuvers such as 10 consecutive somersaults in 11 seconds while maintaining stability amid wind disturbances. As Kevin Chen highlighted, this approach achieves flight performance comparable to real insects in speed, acceleration, and pitching angle, marking a new era in miniature robot autonomy and dexterity.

In a groundbreaking development from Graz University of Technology, researchers have engineered a brain-inspired AI model that mimics three core neural mechanisms—cognitive maps, stochastic neural computations, and compositional coding—to achieve flexible planning and problem-solving with drastically reduced energy consumption. Mirroring the human brain's remarkable efficiency of operating on roughly 20 watts, this AI circumvents exhaustive calculations by efficiently conceiving and testing solutions, marking a significant departure from the energy-intensive demands of traditional large AI systems. Wolfgang Maass encapsulates this paradigm shift, stating, 'our work shows that powerful AI does not necessarily require huge data centers and enormous amounts of energy,' signaling a new era where AI's prowess is decoupled from its carbon footprint.

Beyond energy savings, the Graz team's model excels in adaptability, capable of responding flexibly to novel or altered environments without the need for retraining—a feature particularly vital for autonomous robots and edge devices operating under strict energy constraints. As Yukun Yang emphasizes, 'our AI model can also react flexibly to changed or new situations without needing to be retrained,' underscoring its suitability for real-world applications where continuous learning on the fly is impractical. This flexibility not only enhances operational autonomy but also positions brain-inspired AI as a compelling alternative to large language models, which often rely on massive datasets and centralized computing resources.

Graz University’s brain-inspired AI model delivers flexible, retraining-free planning on just 20 watts, signaling a shift from energy-hungry AI toward adaptive intelligence that thrives at the edge.