Over the course of the quarter I have been conducting a literature review on localization techniques that can be used for the lidar localization investigation. Progress in this area was slow at first was it was difficult to differentiate between the myriad of localization algorithms given different hardware and computational constraints. The break through came when the term range-only SLAM was found. This particular derivative of SLAM is based on the more difficult problem of performing localization and mapping when only range information is available.

The papers listed below address this problem under the assumption that information is available from transmitting beacons from which time of flight calculations can be performed in order to deduce ranging estimates.

- Kantor, G. and Singh, S., 2002. Preliminary results in range-only localization and mapping
- Newman, P. and Leonard, J., 2003, September. Pure range-only sub-sea SLAM
- Martinson, E.B. and Dellaert, F., 2003, September. Marco Polo localization
- Blanco, J.L., González, J. and Fernández-Madrigal, J.A., 2008, May. A pure probabilistic approach to range-only SLAM
- Blanco, J.L., Fernández-Madrigal, J.A. and González, J., 2008, September. Efficient probabilistic range-only SLAM

The scenarios which the papers above describe are different to that of the scenario we are trying to solve, however there are parallels which can be drawn and insight which can be derived from the solutions that were implemented. In the beacon transmitting case there is access to well defined information regarding the locality of the beacons and the complexity of the problem is in mitigating the effect of poor measurement readings.

In our scenario the lack of beacons can be supplemented by identifying key points which can be used as proxies for beacons such that ranging can be performed according to these points and correlated to earlier lidar sweeps.

Lidar localization can therefore be framed as a compounded range-only localization scenario whereby key points first need to be identified and subsequently localization can be performed in the same manner as the beacon present implementations.

The novelty of this work will therefore be centered around creating a key point identifier and correlator given the 2D sweep profile.

Other papers which gave insight into the problem included:

This paper did not use transmitting beacons as was the case the the above mentioned papers but rather assumed that a sonar sensor with an aperture of 2π was used. This assumption further complicates the matter since only one reading is available which corresponds the the distance to the closest object in the environment. The solution was an offline iterative algorithm with a high computational complexity that would be infeasible for this scenario context.

Lastly it is noteworthy to mention that range-only SLAM is not the only form of measurement deficient SLAM. For example in the case of a monocular camera the range to an object is not available but the bearing to the obstacle is. Bearing-only SLAM techniques have their own focus within the literature but topic out of the scope of this investigation.