Due to our team’s recent breakthroughs in physical AI and AIoT, this website and the websites of all 64 portfolio companies are being substantially upgraded to incorporate our Physical AI and AIoT engines and sensing technologies. This overhaul will be completed by Nov. 30, 2026.
This page has not been updated yet.

Distributed Intelligence, Robotics and Physical Logistics

This research program is conducted by a team of Ph.D. researchers with expertise in robots, drones and autonomous vehicles.

Develop intelligence for industrial robots, drones and autonomous vehicles that perceive their surroundings, make bounded decisions and act on physical systems. Research covers robotic manipulation, aerial inspection, autonomous mobility and coordination across connected platforms. Selected projects connect these capabilities to the Aperture Physical AI Engine™ through authorized action and verified feedback.

Resilient Edge AI and Distributed Industrial Intelligence

Allocate inference and decision tasks across sensors, gateways, local servers and cloud services. Investigate model compression, event driven communication and stale state detection. Define which functions must remain local and how queued events reconcile after an outage.

Research questions

When should a task move between nodes? How can recovery prevent a delayed command from being executed after its operational context has changed?

Proposed Demonstration

Run several edge nodes on a small process while limiting bandwidth, removing cloud access and overloading one node.

Evaluation

Latency percentiles, bandwidth, compute and energy use, retained functionality, stale command rejection and recovery time.

Multi Agent Coordination for Industrial Operations

Coordinate agents responsible for inventory, inspection, maintenance and production using explicit responsibilities and shared resource rules. Compare centralized scheduling with distributed negotiation. Investigate conflict resolution, bounded retries, deadlock detection and uncertainty in shared state.

Research questions

When does distributed coordination improve resilience? How should agents resolve competing goals when no plan satisfies every request?

Proposed Demonstration

Coordinate replenishment, inspection and maintenance across several stations; remove one agent and introduce contradictory state updates.

Evaluation

Task completion, resource utilization, deadlocks, conflicting actions, communication overhead and recovery from agent failure.

AI Assisted Material Handling Warehouse Robotics and Physical Supply Chain Execution

Use RFID and UWB context alongside robot perception to assign loads, routes and delivery tasks. Study uncertain inventory, blocked aisles, changing destinations and transfer verification. Selected experiments can connect identification and AI planning to the Aperture Physical AI Engine™ action and feedback pathway.

Research questions

How should uncertain load identity affect dispatch? What evidence confirms that the intended material reached the correct destination?

Proposed Demonstration

Use an existing mobile platform to move identified materials between stations with blocked routes and failed pickup or handoff scenarios.

Evaluation

Delivery accuracy, travel distance, throughput, congestion, interventions and verified transfer completion.

Adaptive Robotic Assembly and Machine Tool Assistance

Adapt one task such as insertion, fastening, loading or a permitted machining adjustment to variations in part geometry, pose and tool condition. Combine vision and force sensing with bounded motion policies and failure detection. Require suitable equipment and process expertise.

Research questions

Can the system identify the difference between misalignment and a damaged part? Which recoveries are repeatable within the approved force and motion envelope?

Proposed Demonstration

Perform a repeatable assembly or tending task with varied part positions, tolerances and controlled obstructions.

Evaluation

Success rate, cycle time, dimensional quality, contact force limits, setup effort and recovery success.

Mobile Robotic Inspection and Active Survey Planning

Plan inspection routes and viewpoints for mobile robots or appropriately operated drones around industrial assets. Investigate coverage, localization drift, battery limits and incomplete observations. Keep navigation feasibility distinct from defect recognition so failures can be diagnosed separately.

Research questions

When should a platform revisit an uncertain observation? Can it preserve useful coverage after a route becomes unavailable?

Proposed Demonstration

Inspect a mock equipment area using an existing mobile platform, comparing fixed routes with uncertainty guided revisits.

Evaluation

Inspection coverage, missed regions, localization error, defect evidence quality, mission time, energy use and operator interventions.

Robot Learning Adaptive Manipulation and Task Transfer

Investigate how existing robot learning methods, imitation learning and vision language action models can support bounded manipulation tasks. Combine visual observations, force or torque feedback and task constraints to adapt grasping, insertion and tool use. Evaluate transfer between objects and workstations without assuming unrestricted general purpose autonomy.

Research questions

Which demonstrations improve generalization to unfamiliar parts? Can the robot identify when an object, contact condition or instruction is outside its validated task range?

Proposed Demonstration

Use an existing robot arm for picking or insertion with varied objects and poses. Compare a fixed task policy, a learned policy and a learned policy with independent action constraints.

Evaluation

Task success on unseen objects, demonstration effort, force limit violations, grasp failures, intervention rate and recovery effectiveness.

Human Robot Collaboration and Shared Workspaces

Study how industrial robots and autonomous mobile robots interpret human movement, task intent and shared workspace occupancy. Investigate handovers, work zone allocation, uncertainty aware motion and clear communication of planned actions. Keep independent protective functions and emergency stops separate from experimental AI behavior.

Research questions

How should uncertain human intent change robot speed or route selection? Can operators understand an approaching handover and interrupt it without confusion?

Proposed Demonstration

Evaluate simulated shared tasks before a controlled low energy handover or shared aisle demonstration using an existing platform and appropriate supervision.

Evaluation

Separation margin violations, near conflict events, task completion, interruption response, operator workload and recovery after an unexpected human movement.

Autonomous Drone Perception Localization and Obstacle Avoidance

Develop aerial perception and navigation methods for industrial facilities, construction sites and infrastructure corridors. Combine cameras, depth sensing, inertial measurements and available positioning signals. Study degraded GNSS, visual ambiguity, changing light, moving obstacles and landing area assessment while respecting platform operating limits.

Research questions

How can a drone detect localization drift before its route becomes unreliable? Which observations justify continued flight, a revised route or a controlled termination of the mission?

Proposed Demonstration

Begin in simulation, then use a protected test area or suitable controlled flight environment with an existing drone. Introduce visual occlusion and positioning uncertainty without radiating interference.

Evaluation

Localization error, obstacle detection coverage, route deviation, collision free mission completion, uncertainty calibration and successful fallback or landing.

Drone Inspection Mission Planning and Adaptive Aerial Sensing

Plan aerial surveys around inspection priorities, asset geometry, observation quality, energy reserves and changing site conditions. Study adaptive viewpoints, anomaly triggered revisits and repeatable comparisons across missions. Applications include bridges, power infrastructure, pipelines, stockpiles and construction progress. Link inspection findings to the correct asset and maintenance record.

Research questions

When is another view worth its flight time? Can a mission distinguish apparent change caused by viewpoint or weather from an asset condition that needs inspection?

Proposed Demonstration

Compare fixed and adaptive inspection routes over a mock asset or validated digital environment, using an existing drone only where suitable operations are available.

Evaluation

Inspection coverage, image suitability, missed defects, repeatability of measurements, mission time, energy reserve and correct association of observations with assets.

Autonomous Industrial Vehicle Perception Planning and Motion Control

Study localization, traversability, obstacle detection, trajectory planning and bounded motion control for yard tractors, mining vehicles, agricultural platforms and other industrial autonomous vehicles. Account for load, stopping distance, terrain, traction, visibility and interaction with workers or manually operated vehicles. Define the conditions within which each experimental policy may operate.

Research questions

How should uncertain terrain or an occluded intersection change a vehicle’s motion plan? Can a planner remain feasible when load and available stopping distance change?

Proposed Demonstration

Use vehicle simulation followed by a scaled platform or controlled industrial test area. Compare planning under nominal conditions with degraded sensing, blocked routes and changing loads.

Evaluation

Localization and tracking error, obstacle misses, stopping margin, route completion, motion constraint violations and appropriate requests for human intervention.

Mixed Fleet Coordination for Robots Drones and Autonomous Vehicles

Coordinate task assignment, route reservations, charging, inspection handovers and material transfers across heterogeneous platforms. Study shared maps, resource conflicts, communication delays and failure recovery. Keep fleet level task decisions separate from each platform’s local motion and protective controls.

Research questions

When does coordination outperform independent operation? How can the fleet reassign a failed drone inspection or vehicle delivery without duplicating tasks or creating a traffic conflict?

Proposed Demonstration

Coordinate simulated aerial inspection, ground inspection and material delivery missions, then connect available platforms to a shared dispatch interface.

Evaluation

Fleet throughput, task duplication, deadlocks, charging availability, missed handovers, local autonomy during outages and recovery from a failed platform.

Explore the Other Programs