example of using isaac with reachy

Signed-off-by: Timothy Le <tle@nebius.com>

Author: Timothy Le

applications/osmo/workflows/osmo/test_isaac_reachy.yaml

@@ -0,0 +1,83 @@
+workflow:
+  name: osmo-isaac-reachy-rl
+  pool: default
+
+  resources:
+    nebius-cli:
+      cpu: 1
+      memory: 2Gi
+      platform: L40S
+
+  tasks:
+    - name: trigger-reachy-training
+      image: ubuntu:22.04
+      resource: nebius-cli
+
+      environment:
+        SUBNET_ID: "vpcsubnet-XXX"
+        PARENT_ID: "project-XXX"
+        FILESYSTEM_ID: "computefilesystem-XXX"
+        NEBIUS_IAM_TOKEN: "NEBIUS_IAM_TOKEN"
+        NGC_API_KEY: "nvapi-XXX"
+        # RL training parameters
+        TASK: "Isaac-Reach-Franka-v0"  # swap for Isaac-Reach-Reachy-v0 when extension is installed
+        NUM_ENVS: "4096"
+        MAX_ITERATIONS: "500"
+
+      command:
+        - bash
+        - -lc
+      args:
+        - |
+          set -euo pipefail
+
+          echo "--- 1. Install Nebius CLI ---"
+          apt-get update && apt-get install -y curl bash jq
+
+          curl -sSL https://storage.eu-north1.nebius.cloud/cli/install.sh | bash
+
+          export PATH="$PATH:/root/.nebius/bin"
+          nebius version
+
+          echo "--- 2. Configure Nebius credentials ---"
+          mkdir -p /root/.nebius
+          printf 'current-profile: default\nprofiles:\n  default:\n    endpoint: api.eu-north1.nebius.cloud:443\n    parent-id: %s\n' \
+            "$PARENT_ID" > /root/.nebius/config.yaml
+
+          echo "--- 3. Launch Isaac Lab training job ---"
+          RUN_NAME="${TASK}-$(date +%s)"
+          JOB_OUTPUT=$(nebius ai job create \
+            --name "reachy-train-$(date +%s)" \
+            --image "nvcr.io/nvidia/isaac-lab:2.3.2" \
+            --registry-username "\$oauthtoken" \
+            --registry-password "$NGC_API_KEY" \
+            --platform "gpu-l40s-d" \
+            --preset "1gpu-16vcpu-96gb" \
+            --volume "$FILESYSTEM_ID:/mnt/storage" \
+            --container-command "bash" \
+            --args "-c '/workspace/isaaclab/isaaclab.sh -p /workspace/isaaclab/scripts/reinforcement_learning/rsl_rl/train.py --task $TASK --headless --num_envs $NUM_ENVS --max_iterations $MAX_ITERATIONS +runner.log_root_path=/mnt/storage/logs 2>&1'" \
+            --timeout "120m" \
+            --subnet-id "$SUBNET_ID" \
+            --parent-id "$PARENT_ID")
+          echo "Job create response: $JOB_OUTPUT"
+          JOB_ID=$(echo "$JOB_OUTPUT" | grep -oE 'aijob-[a-z0-9]+' | head -1)
+          echo "Submitted training job: $JOB_ID"
+
+          echo "--- 4. Wait for training completion ---"
+          while true; do
+            JOB_RAW=$(nebius ai job get --id "$JOB_ID" --format json 2>&1)
+            STATUS=$(echo "$JOB_RAW" | jq -r '.status.phase // .status.state // empty' 2>/dev/null \
+              || echo "$JOB_RAW" | grep -oE '(RUNNING|SUCCEEDED|COMPLETED|FAILED|ERROR|CANCELLED|STARTING|PENDING)' | head -1)
+            echo "Training job status: $STATUS"
+            case "$STATUS" in
+              SUCCEEDED|COMPLETED)
+                echo "Training completed successfully."
+                echo "Logs and checkpoints saved to: /mnt/storage/logs/$TASK/"
+                break ;;
+              FAILED|ERROR|CANCELLED)
+                echo "Training failed with status: $STATUS"
+                nebius ai job logs "$JOB_ID" 2>&1 | tail -50
+                exit 1 ;;
+              *) sleep 60 ;;
+            esac
+          done