Sensor fusion and mapping
Evaluate LiDAR, inertial, radar and visual or thermal inputs where appropriate to the platform and environment. Reject inconsistent observations and validate place matches before using them to correct drift.
Initiatives · Subterra
Subterra develops onboard positioning, mapping and navigation integrity for autonomous machines in unfamiliar, GPS-denied underground environments.
Current statusIn development
Make navigation uncertainty visible to motion planning so movement can be constrained when localisation becomes unreliable.
Subterra develops navigation methods for mines, tunnels, shafts and caves. Its scope covers local positioning and mapping, bounded onboard planning and coordinated exploration by multiple robots without continuous cloud connectivity.
Darkness, dust, repeated geometry, vibration and sensor faults can degrade localisation or create a convincing but incorrect map match. A machine needs to detect weak or inconsistent observations as well as estimate its position.
Combine platform-appropriate range, inertial and visual observations through simultaneous localisation and mapping (SLAM). Check timing, calibration, sensor health and geometric consistency; assess the estimate’s uncertainty before supplying it to a planner. Map corrections must trigger controlled replanning while continuous motion estimates remain usable by the controller.
Mining and infrastructure operators, underground inspection and emergency reconnaissance teams, and robotics developers evaluating navigation under degraded sensing and communications. Civilian and defence engineering applications are assessed against their own platforms, operating permissions and field conditions.
Evaluate LiDAR, inertial, radar and visual or thermal inputs where appropriate to the platform and environment. Reject inconsistent observations and validate place matches before using them to correct drift.
Assess whether reported uncertainty reflects observed error and detect conditions that invalidate the estimate. Platform-defined limits govern slowing, holding or recovery; covariance alone does not establish safe navigation.
Target local estimation and bounded planning during communications loss. Measure latency, power and thermal demands, and retain evidence for reconciliation after reconnection. Recovery must account for energy and reachable return routes.
ROS 2-compatible interfaces are a development target, with explicit frames, timestamps and stale-data handling. Each platform retains its motion, collision and recovery constraints.
Research shared observations and task allocation across robots. Validate map alignment and avoid double-counting shared evidence; account for communications loss, energy reserves and reachable recovery routes.
In a repetitive tunnel, similar surfaces may produce a false place match. The intended response is to challenge the match against independent motion evidence, restrict movement when the estimate is unreliable and seek a better observation or recovery route. Platform testing must establish when that response is effective.
Evaluate trajectory and map error, uncertainty calibration, failure detection and recovery against independent reference measurements where practicable. Replay and simulation support field trials under degraded sensing and communications. Multi-robot evaluation adds map consistency, task completion and safe recovery.
Begin with one robot, sensor configuration and representative underground environment. Establish the baseline and controller responsibilities before evaluating integrated behaviour. Navigation performance and safety acceptance belong to the complete platform and operating case.