В MIT создали чип, который даёт крошечным роботам возможность строить 3D-карты на ходу

Researchers from MIT presented a new chip capable of turning tiny robots into fully autonomous navigators. The device combines a specialized hardware module and an optimized algorithm — together they build 3D maps of the environment in real time with minimal memory and energy consumption.

Challenge: small robot, complex environment

Small robots attract engineers for an obvious reason: they can squeeze into places where humans or large machines cannot go. But they face a fundamental limitation — a compact body can accommodate only a miniature battery and low-power chip. Classical navigation algorithms, such as SLAM (Simultaneous Localization and Mapping), require significant computation. Building a three-dimensional map of an unfamiliar space while simultaneously tracking one's own position in it — a task that in its standard implementation consumes gigabytes of memory and several watts of power. For a device the size of a beetle, this is overwhelming.

Furthermore, standard SLAM implementations store maps as dense three-dimensional arrays: the volume of data grows rapidly with the size of the explored space, making them fundamentally inapplicable in embedded systems with limited memory. The standard workaround — transmitting raw sensor data to an external computer, receiving a finished map back. This works in a laboratory, but fails where there is no connection: under rubble, in a tight pipeline, inside the human body.

Algorithm and hardware as a unified solution

The MIT team applied a hardware-software co-design approach: the mapping algorithm and hardware accelerator were developed together from scratch. The result is a chip where every operation is optimized for specific hardware, rather than for a universal processor.

• The algorithm was rewritten to minimize memory accesses and uses a compact incremental map representation
• The hardware module processes the most resource-intensive stages in parallel
• A 3D map is built directly onboard, without transmitting data externally
• Energy consumption is reduced to a level suitable for tiny batteries

This approach is not new in microelectronics as a whole, but it is rarely applied to robotics navigation tasks. Most teams either improve the algorithm or create specialized hardware. Doing both for a single task is more difficult, but it is precisely this that yields a multiple gain in efficiency. The MIT researchers presented details of the architecture in an academic publication, though specific parameters for accuracy and speed on benchmark scenarios have not yet been disclosed publicly.

Where a map without a server is needed

There are several practical applications — they are united by one characteristic: environments where there is no connection to the operator and no room for cables.

• Search and rescue operations — robots under rubble after earthquakes, where there is no GPS and no stable radio signal
• Medicine — surgical probes and targeted drug delivery systems operating inside the body without a control cable
• Industrial inspection — pipelines, air ducts, reactor cavities inaccessible to humans
• Agriculture — mini-drones for crop monitoring or pollination in enclosed greenhouses

Particular interest lies in medical applications: autonomous probes that do not require constant cable-based control can significantly simplify minimally invasive surgical procedures. In all other cases, the key capability is one — to make navigation decisions independently, without external computing infrastructure.

What this means

The MIT chip eliminates a compromise that the robotics industry has long tolerated: either a large apparatus with a powerful onboard computer, or a compact one that depends on external infrastructure. If a similar module becomes part of the standard toolkit for developers, tiny autonomous devices will gain access to environments that were previously physically closed to them.