Researchers at the Massachusetts Institute of Technology (MIT) have developed a new system-on-a-chip designed to enable small, battery-powered robots and devices to generate detailed three-dimensional maps of their surroundings in real time while using extremely low power. The chip consumes approximately 6 milliwatts—comparable to the power usage of a single LED—making it especially suited for use in tiny drones, unmanned aerial vehicles (UAVs), and augmented reality headsets.
What Happened
The MIT team introduced the chip, named Gleanmer, which integrates an innovative mapping algorithm called GMMap with specialized hardware to efficiently process 3D environmental data. Unlike conventional mapping techniques that rely on cubic volumetric pixels (voxels) and require substantial memory and power, Gleanmer employs ellipsoid representations known as Gaussians to capture obstacle shapes more compactly and flexibly.
This approach reduces both the memory footprint and computational energy demand, enabling the chip to operate on roughly 2.5% of the power typically needed by competing map construction systems. The researchers demonstrated Gleanmer reconstructing complex 3D environments from live data, including footage streamed directly from an iPhone camera, while maintaining real-time performance.
The work was presented recently at the IEEE Very Large-Scale Integrated Circuits Symposium and benefited from support by Intel, the U.S. National Science Foundation, Amazon, and the MIT-MathWorks Fellowship.
Key Facts
MIT’s chip is designed to function in power-constrained settings such as UAVs inspecting industrial HVAC systems for gas leaks or AR headsets used in medical training and equipment repair. The system’s primary innovation is the use of ellipsoidal Gaussians instead of voxels, which drastically cuts down the data size required to represent the environment by allowing smooth adaptation to curved object surfaces.
The mapping algorithm processes input depth images in a single pass, avoiding the need to store entire images and reducing memory requirements significantly. Additionally, the hardware co-design keeps frequently accessed Gaussian data in fast on-chip memory, further minimizing power-heavy off-chip memory fetches.
The chip operates at about 6 milliwatts, using only around 20% of the energy normally required for path planning in autonomous navigation tasks.
What This Means
Gleanmer’s ability to generate precise, real-time 3D maps at ultra-low power consumption marks a significant advancement for tiny autonomous systems that have traditionally been limited by battery life and computing capacity. Small drones and robots can now potentially perform complex navigation in tight or hazardous environments with longer operational times without frequent recharging.
For augmented reality devices, this chip could drastically improve user experience by enabling lightweight, untethered headsets that maintain continuous spatial awareness without compromising battery longevity or requiring bulky hardware.
This technology also illustrates the importance of algorithm and hardware co-design, showing how integrating efficient software methods with custom silicon accelerators can unlock new capabilities in low-power edge computing. Such innovation is critical as demand grows for smart devices that operate independently in real-world settings.
Background
Traditional 3D mapping systems rely heavily on voxels to track obstacles and free space, which demands large memory and high-energy consumption unsuitable for small devices. MIT’s use of Gaussian ellipsoids in its GMMap algorithm was a breakthrough step to compactly and flexibly model environmental features while conserving resources.
Prior to Gleanmer, generating detailed obstacle maps typically required multiple passes over high-resolution depth images and significant data storage, limiting deployment in battery-constrained edge devices.
What Remains Unclear
The reviewed sources do not specify the commercial availability timeline for Gleanmer or detailed integration plans with specific vendor products. It is also not confirmed how the chip performs outside laboratory environments or under varying operational conditions such as outdoor deployments.
What Comes Next
The researchers intend to increase energy efficiency further by relocating processing units closer to sensors on the chip. They are exploring expanding the use of Gaussian representations for other applications, such as enhancing AI systems’ ability to understand complex schematics and blueprints efficiently.
Sources
This article is based on reporting and publicly available information from the following sources:
Read more Science & Technology stories on Goka World News.