v0.1.6

Major features and improvements:

• Renamed Habitat-API to Habitat-Lab to emphasize its purpose: to conduct experiments in embodied AI, define tasks, train different agents, benchmark performance (#448, #450).
• Implemented a demo cubemap2equirect that stitches together 6 sensors into one creating a 360 Sensor (#478).
• Added an observation transformer registry for transformation as Cropping, Resize and separating Observation Transformation and Policy (#478).
• Embodied Question Answering (EQA) CNN Pre-trainer for EQA baseline implementation by @mukulkhanna (#430).
• Add UUID config to simulator visual sensors to better support multiple sensor of the same type (#472).
• Implemented standard interface for policy instantiation by @rpartsey (#453).
• Added Habitat-Lab Colab Tutorial, presented on ECCV2020 by @dexter1691(#445).
• Delivered Habitat-Lab Interactive Task Colab Tutorial that demonstrate adding new task with grab/release action and moving objects by @dexter1691 (#449).
• Implemented DDPPO baseline for Image Navigation task by @thibautlavril (#378).
• Implemented policy registry to support different policies for the same Trainer by @rpartsey (#440).
• Added usage habitat-sim's topdown view (#442).
• Implemented dataset generation script for Gibson 500 scenes PointNav Task by @srama2512 (#387).
• Removed Habitat Sim wrapper to avoid sim._sim constructions. (#418).

Other improvements and bug fixes:

• Bug fix for VLN follower (#461)
• Updated precommit and fix linter (#433)
• Added isort to jupytext (#441)
• Added autoflake support (#456)
• Flake8 upgrade and Pre-Commit Hook added for CircleCI (#459, #469)
• Multiple other Fixes: signal masking, README updates (#406, #413, #418, #435, #426, #438, #439)

Corresponding Habitat-Sim v0.1.6 release.

Kudos to our community contributors: @thibautlavril, @rpartsey, @dexter1691, @mukulkhanna, @srama2512.

v0.1.5

Major features and improvements:

• Added Object Navigation task definition, dataset, metrics, goal sensor (#284, #309, #355)
• Object Navigation DDPPO baselines (#370)
• Added Image navigation task (#333)
• Added geodesic distance for multiple goals (#290)
• Integrated RGB, Depth, PyRobot actuation noises, initial camera orientation. (#305)
• Generated PointNav Gibson v2 supporting sim2real direction (#316)
• Nightly build of docs and tutorials (#393, #396, #400)
• Faster multi-goal geodesic distance computation and caching (#352)

Other improvements and bug fixes:

• Metrics dependency mechanism to deliver ObjectNav 2x speed (#284, #332)
• Cleaned up agent state checks (#399)
• Added utility func for computing local state change (#392)
• Fixed docker build with extra space. (#363)
• Update demo notebook (habitat.utils.common import) (#351)
• Update Dockerfile to stable (#339)
• Update CUDAGL base image, cmake version and git clone in Dockerfile. (#349)
• Added physics config to default habitat simulator config. (#289)
• Interface with habitat-sim no sliding (#283)
• Added Observation Transformation API (#365)
• Soft SPL implementation: SPL with a softer success criterion (#361).
• Habitat Challenge 2020 gRPC evaluation support (#319)

Corresponding Habitat-SIM v0.1.5 release.

Kudos to our community contributors: @thibautlavril, @Xiaoming-Zhao, @pushkalkatara, @kdexd, @Adidushi, @smorad.

v0.1.4

• A new large-scale distributed reinforcement learning algorithm Decentralized Distributed PPO (#245)
• Added Visual Language Navigation Task & Dataset with shortest path example and tests (#233, #263)
• Allow free form arbitrary yaml configs (#275)
• Added test coverage measure for habitat-api with codecov (#257)
• Expand Tensorboard visualization to more than RGB sensor (#251)

Other improvements and bug fixes:

• Preserve scene ordering (#234)
• Fix bugs in config setup and pause_env condition (#235)
• Add CUDA path to CI config path (#249)
• Fix wrong return type of SimulatorTaskAction.step (#248)
• Fix try register for pointnav dataset (#254)
• Fix the documentation build (#256)
• Fix contains method in space classes (#272)
• Fixing Spelling error in 'appned' and adding test (#280)

Corresponding Habitat-SIM v0.1.4 release.

Community contributors: @koshyanand, @jacobkrantz, @arjunmajum, @ChanganVR, @danmou.

v0.1.3

• Habitat baselines redesign: introduced general trainer (#132, #153, #172, #178), PPO baseline is extendable to other tasks (#112, #126, #144, #176), customizable dataset iterator (#163, #156, #216)
• Embodied Question Answering task released with a dataset. (#113)
• Towards sim2real: integration with PyRobot (#199)
• Continuous and parameterized actions support, extendable simulator action space (#110, #196, #151)
• GPU-GPU memory sharing with PyTorch to speed up training (#197)
• Visualization tools: tensorboard integration (#127), display an episode navigation video on tensorboard, fog of war (#160, #166)
• Added documentation generating system (#201, #219, #212) and improved tutorials (#125, #128), updated class diagram (#224)

Other improvements and bug fixes:

• CI improvements: Added ccache, fixed PyTorch caching (#161), install hsim requirements step (#159), removed travis CI (#215), check of test data installed (#193), pre-commit, added isort-seed, changed CI black and isort (#214)
• Vector environments bug fixes: fix multiprocess eval (#150), fixed vector env deletion (#185), fixed pausing logic (#200), fixed bug in env_utils (#208), fixed a bug of empty list initialization (#209)
• Dataset utils fixed: modified partitioning logic to handle left-over scenes (#89), fixed bug with uneven splits for dataset (#192).
• Correct the naming of PointGoalNav Sensors (#180), fixed an issue with POINTNAVV1 name (#130)
• Fix quickstart keys commands (#147)
• Update Habitat-API to allow for no rendering sensors (#139)
• Update baseline test import and fix docstring (#170)
• Evaluate on val and add an option to not take the checkpoints config (#213)
• Changed Shortest Path Follower criteria to use geodesic distance. (#194)
• Add py.typed file to indicate PEP 561 compatibility (#198)

Corresponding Habitat-SIM v0.1.3 release.

Community contributors: @danielgordon10, @RicCu, @danmou

v0.1.2

• Added SLAM based agents
• Added Greedy Shortest Path Follower
• Added episode generation code for PointGoal task.
• Nav task improvements, static PointGoal sensor
• Image to video generation utils and top down map measure
• Dataset utility functions
• Config system improvements: compositional config, added scenes_dir.
• Made simple agents suitable for challenge submissions
• Redesign factory/registry pattern
• Switched to attr library for class definition
• Separate baseline installation options
• CI setup with e2e testing using GPU machines
• Tutorials, examples and improved documentation
• Numerous number of bug fixes

Community contributors: @danielgordon10, @RicCu.

v0.1.1

The release contains important parts for the Habitat challenge, visualizations utilities, fixes:

• Benchmark and challenge implementations
• PPO agents, Random, ForwardOnly, GoalFollower agents
• High resolution top down map and bird-eye visualizations
• Collisions measure introduced
• SPL success criteria fix
• Added deleter that closes subprocesses

Community contributors: @ducha-aiki, @danielgordon10, @titardrew, @eric-xw, @oawiles, @mjschock.

v0.1.0

Habitat-API is a modular high-level library for end-to-end development in embodied AI. It supports the following functionalities:

• Defining embodied AI tasks (e.g. navigation, instruction following, question answering)
• Configuring embodied agents (physical form, sensors, capabilities)
• Training these agents (via imitation or reinforcement learning, or no learning at all as in classical SLAM)
• Benchmarking their performance on the defined tasks using standard metrics.