Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
Expand Up @@ -5,16 +5,20 @@ Gym ID: `ANYmalCFlatRL-v1`.

| Deployment | Environment | PPO configuration |
|---|---|---|
| `newton` (DexUni, recommended) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |
| `default` (Default) | [env.yaml](env.yaml) | [agents/ppo.yaml](agents/ppo.yaml) |
| `newton` (Newton) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |

Each PPO file's `trainer.gym_config` selects its matching deployment. The
catalog defaults to `default`; backend selection belongs to the environment file.
catalog defaults to `newton`; backend selection belongs to the environment file.
The Newton deployment uses DexUni with dynamic joints and automatic contact
selection. For this rigid articulation scene, DexUni dispatches to MuJoCo Warp
on CUDA (`use_mujoco_cpu: false`). Its MuJoCo settings are explicit, including
20 solver iterations and 50 line-search iterations. Both deployments preserve
unbounded policy actions with `clip: null`.

```bash
embodichain show-task embodichain_tasks:anymal_c_flat
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/anymal_c_flat/env.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/anymal_c_flat/agents/ppo.yaml
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/anymal_c_flat/env.newton.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/anymal_c_flat/agents/ppo.newton.yaml
```

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ tags:
- locomotion
- reinforcement_learning
- quadruped
default_deployment: default
default_deployment: newton
deployments:
default:
config: env.yaml
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -14,14 +14,21 @@ physics_config:
- -9.81
num_substeps: 1
solver_cfg:
solver_type: mujoco_warp
njmax: 1024
nconmax: 256
iterations: 10
ls_iterations: 50
integrator: implicitfast
separate_worlds: true
tolerance: 1.0e-08
solver_type: dexuni
joint_mode: dynamic
contact_mode: auto
mujoco_options:
njmax: 1024
nconmax: 256
iterations: 20
ls_iterations: 50
integrator: implicitfast
separate_worlds: true
tolerance: 1.0e-08
cone: pyramidal
impratio: 1.0
ls_parallel: false
use_mujoco_cpu: false
physics_dt: 0.005
env:
sim_steps_per_control: 4
Expand Down Expand Up @@ -52,7 +59,8 @@ env:
actions:
joint_position:
contract: joint_position.default_offset@1
params: {}
params:
clip: null
extensions: {}
robot:
class_type: ANYmalC
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,8 @@ env:
actions:
joint_position:
contract: joint_position.default_offset@1
params: {}
params:
clip: null
extensions: {}
robot:
class_type: ANYmalC
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -5,16 +5,20 @@ Gym ID: `UnitreeG1FlatRL-v1`.

| Deployment | Environment | PPO configuration |
|---|---|---|
| `newton` (DexUni, recommended) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |
| `default` (Default) | [env.yaml](env.yaml) | [agents/ppo.yaml](agents/ppo.yaml) |
| `newton` (Newton) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |

Each PPO file's `trainer.gym_config` selects its matching deployment. The
catalog defaults to `default`; backend selection belongs to the environment file.
catalog defaults to `newton`; backend selection belongs to the environment file.
The Newton deployment uses DexUni with dynamic joints and automatic contact
selection. For this rigid articulation scene, DexUni dispatches to MuJoCo Warp
on CUDA (`use_mujoco_cpu: false`). Its MuJoCo settings are explicit, including
20 solver iterations and 50 line-search iterations. Both deployments preserve
unbounded policy actions with `clip: null`.

```bash
embodichain show-task embodichain_tasks:g1_flat
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/g1_flat/env.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/g1_flat/agents/ppo.yaml
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/g1_flat/env.newton.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/g1_flat/agents/ppo.newton.yaml
```

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ tags:
- locomotion
- reinforcement_learning
- humanoid
default_deployment: default
default_deployment: newton
deployments:
default:
config: env.yaml
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -15,14 +15,21 @@ physics_config:
- -9.81
num_substeps: 1
solver_cfg:
solver_type: mujoco_warp
njmax: 1024
nconmax: 256
iterations: 20
ls_iterations: 50
integrator: implicitfast
separate_worlds: true
tolerance: 1.0e-08
solver_type: dexuni
joint_mode: dynamic
contact_mode: auto
mujoco_options:
njmax: 1024
nconmax: 256
iterations: 20
ls_iterations: 50
integrator: implicitfast
separate_worlds: true
tolerance: 1.0e-08
cone: pyramidal
impratio: 1.0
ls_parallel: false
use_mujoco_cpu: false
physics_dt: 0.005
env:
sim_steps_per_control: 4
Expand Down Expand Up @@ -83,7 +90,8 @@ env:
actions:
joint_position:
contract: joint_position.default_offset@1
params: {}
params:
clip: null
extensions: {}
robot:
class_type: UnitreeG1
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,8 @@ env:
actions:
joint_position:
contract: joint_position.default_offset@1
params: {}
params:
clip: null
extensions: {}
robot:
class_type: UnitreeG1
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -5,16 +5,20 @@ Gym ID: `UnitreeGo1FlatRL-v1`.

| Deployment | Environment | PPO configuration |
|---|---|---|
| `newton` (DexUni, recommended) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |
| `default` (Default) | [env.yaml](env.yaml) | [agents/ppo.yaml](agents/ppo.yaml) |
| `newton` (Newton) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |

Each PPO file's `trainer.gym_config` selects its matching deployment. The
catalog defaults to `default`; backend selection belongs to the environment file.
catalog defaults to `newton`; backend selection belongs to the environment file.
The Newton deployment uses DexUni with dynamic joints and automatic contact
selection. For this rigid articulation scene, DexUni dispatches to MuJoCo Warp
on CUDA (`use_mujoco_cpu: false`). Its MuJoCo settings are explicit, including
20 solver iterations and 50 line-search iterations. Both deployments preserve
unbounded policy actions with `clip: null`.

```bash
embodichain show-task embodichain_tasks:go1_flat
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/go1_flat/env.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/go1_flat/agents/ppo.yaml
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/go1_flat/env.newton.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/go1_flat/agents/ppo.newton.yaml
```

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ tags:
- locomotion
- reinforcement_learning
- quadruped
default_deployment: default
default_deployment: newton
deployments:
default:
config: env.yaml
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -15,14 +15,21 @@ physics_config:
- -9.81
num_substeps: 1
solver_cfg:
solver_type: mujoco_warp
njmax: 1024
nconmax: 256
iterations: 10
ls_iterations: 50
integrator: implicitfast
separate_worlds: true
tolerance: 1.0e-08
solver_type: dexuni
joint_mode: dynamic
contact_mode: auto
mujoco_options:
njmax: 1024
nconmax: 256
iterations: 20
ls_iterations: 50
integrator: implicitfast
separate_worlds: true
tolerance: 1.0e-08
cone: pyramidal
impratio: 1.0
ls_parallel: false
use_mujoco_cpu: false
physics_dt: 0.005
env:
sim_steps_per_control: 4
Expand Down Expand Up @@ -83,7 +90,8 @@ env:
actions:
joint_position:
contract: joint_position.default_offset@1
params: {}
params:
clip: null
extensions: {}
robot:
class_type: UnitreeGo1
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,8 @@ env:
actions:
joint_position:
contract: joint_position.default_offset@1
params: {}
params:
clip: null
extensions: {}
robot:
class_type: UnitreeGo1
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -5,16 +5,20 @@ Gym ID: `UnitreeGo2FlatRL-v1`.

| Deployment | Environment | PPO configuration |
|---|---|---|
| `newton` (DexUni, recommended) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |
| `default` (Default) | [env.yaml](env.yaml) | [agents/ppo.yaml](agents/ppo.yaml) |
| `newton` (Newton) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |

Each PPO file's `trainer.gym_config` selects its matching deployment. The
catalog defaults to `default`; backend selection belongs to the environment file.
catalog defaults to `newton`; backend selection belongs to the environment file.
The Newton deployment uses DexUni with dynamic joints and automatic contact
selection. For this rigid articulation scene, DexUni dispatches to MuJoCo Warp
on CUDA (`use_mujoco_cpu: false`). Its MuJoCo settings are explicit, including
20 solver iterations and 50 line-search iterations. Both deployments preserve
unbounded policy actions with `clip: null`.

```bash
embodichain show-task embodichain_tasks:go2_flat
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/go2_flat/env.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/go2_flat/agents/ppo.yaml
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/go2_flat/env.newton.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/go2_flat/agents/ppo.newton.yaml
```

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ tags:
- locomotion
- reinforcement_learning
- quadruped
default_deployment: default
default_deployment: newton
deployments:
default:
config: env.yaml
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -15,14 +15,21 @@ physics_config:
- -9.81
num_substeps: 1
solver_cfg:
solver_type: mujoco_warp
njmax: 1024
nconmax: 256
iterations: 10
ls_iterations: 50
integrator: implicitfast
separate_worlds: true
tolerance: 1.0e-08
solver_type: dexuni
joint_mode: dynamic
contact_mode: auto
mujoco_options:
njmax: 1024
nconmax: 256
iterations: 20
ls_iterations: 50
integrator: implicitfast
separate_worlds: true
tolerance: 1.0e-08
cone: pyramidal
impratio: 1.0
ls_parallel: false
use_mujoco_cpu: false
physics_dt: 0.005
env:
sim_steps_per_control: 4
Expand Down Expand Up @@ -83,7 +90,8 @@ env:
actions:
joint_position:
contract: joint_position.default_offset@1
params: {}
params:
clip: null
extensions: {}
robot:
class_type: UnitreeGo2
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,8 @@ env:
actions:
joint_position:
contract: joint_position.default_offset@1
params: {}
params:
clip: null
extensions: {}
robot:
class_type: UnitreeGo2
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -5,16 +5,20 @@ Gym ID: `UnitreeH1_2FlatRL-v1`.

| Deployment | Environment | PPO configuration |
|---|---|---|
| `newton` (DexUni, recommended) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |
| `default` (Default) | [env.yaml](env.yaml) | [agents/ppo.yaml](agents/ppo.yaml) |
| `newton` (Newton) | [env.newton.yaml](env.newton.yaml) | [agents/ppo.newton.yaml](agents/ppo.newton.yaml) |

Each PPO file's `trainer.gym_config` selects its matching deployment. The
catalog defaults to `default`; backend selection belongs to the environment file.
catalog defaults to `newton`; backend selection belongs to the environment file.
The Newton deployment uses DexUni with dynamic joints and automatic contact
selection. For this rigid articulation scene, DexUni dispatches to MuJoCo Warp
on CUDA (`use_mujoco_cpu: false`). Its MuJoCo settings are explicit, including
20 solver iterations and 50 line-search iterations. Both deployments preserve
unbounded policy actions with `clip: null`.

```bash
embodichain show-task embodichain_tasks:h1_2_flat
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/h1_2_flat/env.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/h1_2_flat/agents/ppo.yaml
embodichain run-task --gym_config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/h1_2_flat/env.newton.yaml
embodichain train-rl --config extensions/embodichain_tasks/configs/tasks/locomotion/velocity/h1_2_flat/agents/ppo.newton.yaml
```

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ tags:
- locomotion
- reinforcement_learning
- humanoid
default_deployment: default
default_deployment: newton
deployments:
default:
config: env.yaml
Expand Down
Loading
Loading