Physics Motion — Human Locomotion
Vladimir Kudryashov
$11.00
$22.00
50%OFF
(no ratings)
Jump AssetStore
Physics Motion Pack — pretrained ML‑Agents models and example agents for physics‑driven humanoid locomotion (idle, walk, run, sprint, backwards). Retargetable to Unity Humanoid rigs.Physics Motion Pack delivers inference‑ready neural motion models and lightweight example controllers that let you add physically simulated humanoid locomotion to Unity projects without training.The package includes multiple small, specialized models (Base / Transition / Turn / Strong) for five locomotion behaviours plus two get‑up models. Models were trained with an ArticulationBody‑based rig for simulation stability; because they target a standard Humanoid mapping, you can retarget the motions to other humanoid characters that follow Unity’s Humanoid/Animator conventions.Included example agents (LocomotionAgent, SimpleLocomotionAgent, SimpleAgent) demonstrate common workflows: switching motions, turning, fall/recovery and applying motion to alternate models via the example Retarget object.Use cases: games, visual previews, demonstration assets, and research/integration where pretrained, stable physics‑based motion is useful. Note: training pipelines are not included in this release.FeaturesMotion models. The package provides pretrained ModelAsset inference models for five locomotion behaviours:IdleWalkingRunSprintWalking BackwardsEach behaviour includes four NN models:Base — high-fidelity imitation of reference animation, minimal jitter.Transition — supports two behaviours and enables transitions between them.Turn — specialized for turning maneuvers.Strong — robust model trained to withstand disturbances; useful for recovery and extended sequences but less natural for transitions/turns.Additionally the package contains two get-up models: GetupFromBack and GetupFromBelly.AgentsLocomotionAgent — full CharacterController-like behaviour: model switching, transitions, turning, strong/fail/get‑up logic.SimpleLocomotionAgent — simplified controller using Strong models (easier setup).SimpleAgent — minimal single-model agent for quick tests.RetargetingModels trained on a Humanoid rig and are retargetable to other humanoid characters. Example scenes include a Retarget GameObject to preview motions on different avatars.ExamplesCharacterController — implemented by LocomotionAgent. Can sequentially switch motions, perform turns, handle falls and get-up behaviours. This is the most feature-complete setup and functions like a CharacterController.SimpleCharacterController — implemented by SimpleLocomotionAgent. Uses only Strong models per behaviour. Easier to configure; motions are less natural.SimpleAgent — implemented by SimpleAgent. Minimal agent controlling a single ModelAsset for quick tests or custom usage.Supported OS Windows, Android, iOS.Tested with Unity 2022.3.62f3 / 2023.2.20f1 / 6000.2.10.f1.DependenciesSentis: 2.1.3.Link to documentationSupport qwegoogl@gmail.com
