Use LAMBKIN for performance benchmarking

[hidmic]

Description

Once we are past #85, we will need extensive performance benchmarking to show beluga_amcl is indeed as good if not better than nav2_amcl. While we could extend #117 to suite all our needs, LAMBKIN may already provide the necessary infrastructure for it -- particularly once Ekumen-OS/lambkin#35 lands.

Definition of done

• Define benchmarks, w.r.t.:
  • Input data
  • Particle filter setup
  • Performance metrics
• Implement benchmarks using LAMBKIN
  • LAMBKIN may require extension
• Publish benchmark reports

[nahueespinosa]

Currently Beluga has been tested on developer laptops. There is no evidence to indicate that it would not work on low-spec platforms, but, as @hidmic pointed out, we should consider running benchmarks in different hardware platforms:

• Raspberry Pi 4B
• Jetson Nano

[hidmic]

A substantial amount of work was pursued by @serraramiro1 already. I'll push it past the finish line.

[hidmic]

We have identified a number of additional datasets we can use for this:

• the Willow Garage dataset;
• the TorWIC dataset;
• the OpenLORIS dataset.

In addition to the Magazino dataset, the MARS dataset, the TUM dataset, and the Rawseeds dataset. CC @glpuga.

[glpuga]

• Hilti Challenge 2022 dataset
• MARS???
• Malaga 2009
• NCTL, through nclt2ros ?
• OpenLORIS
• Rawseeds dataset
• "TUM" (only a subset)
• Willow Garage

[hidmic]

This has materialized in Ekumen-OS/lambkin#93, which will be relocated to this repository with #415 once it consolidates. Pending issues are LAMBKIN specific and we can track them there. Closing as done.