In this section you’ll find information about what’s new in the newer releases of L5Kit.

Release v.1.4.0 (18 Aug 2021)

  • Added L5Kit gym-compatible environment;

  • Added unicycle kinematic model;

  • Added metric sets for closed loop evaluation;

  • Updated the structure and style of L5Kit documentation;

  • Small refactoring of closed loop simulator;

Release v.1.3.1 (8 Jun 2021)

  • Hot-fix to add the visualizer package in the release wheel;

  • Add init to visualizer package;

  • Small fixes in notebooks;

Release v.1.3.0 (19 Apr 2021)

  • Remove black formatter;

  • Add ackerman exact for perturbations, skip applying when the agent is static;

  • Add flag to enable/disable ego history rasterization;

  • Add simulation metrics suite;

  • Add simulation loop;

  • Add new visualizer for improved visualisation based on Bokeh;

  • Add notebooks for simulation;

Release v.1.2.0 (13 Jan 2021)

  • Add an interface for matrix batch transformation;

  • Speed up rasterisation;

  • Improve documentation;

  • Add flip in the rasteriser context (but keep a flag for compatibility);

  • Make agent_sampling modular;

  • Improve MapAPI interface and add lanes interpolation;

  • SatelliteRasteriser for rectangular-shaped has been fixed (thanks pascal-pfeiffer!);

  • Speed up get_frame_indices call;

Release v.1.1.0 (29 Sep 2020)

  • target_positions are now in agent space. See Coordinate Systems and notebooks for examples on how to use it and its relation with the other spaces. This fixes a misalignment which is affecting the competition;

  • Ego and AgentDataset now returns more matrices to convert between spaces;

  • Transformations Matrices notation has changed from “A_to_B” to “B_from_A”;

  • BoxRasterizer now support subpixel precision;

  • Fix an issue with swapped width and height which was preventing to use rectangular-shaped inputs;

  • Introduce Pipenv for deterministic builds;

  • Add option to disable traffic lights;

  • Improve trajectories visualisation params;

Release v.1.0.6 (24 Aug 2020)

  • Add support for multiple metrics;

  • Improve documentation;

  • Improve API for zarr interaction;

  • Bug fixes;

  • Refactor config files;

  • Provide mock-test for competition in prediction notebook;

Release v.1.0.5 (6 Aug 2020)

  • Add first traffic lights implementation;

  • Speed up semantic rasterisation;

  • Introduce MapAPI to handle protobuf;

  • Refactor agents_mask to also include static agents;

  • Improve tests suite;

Release v.1.0.3 (8 Jul 2020)

  • Drop strictyaml;

  • Enforce some deps versions;

Release v.1.0.2 (3 Jul 2020)

  • Speed up training and inference, especially for long scenes;

  • Add better print for dataset;

Release v.1.0.0 (22 Jun 2020)

  • First release of the project;