orby-agent-v0 (#8)

- Add results from Orby Agent (e3f1cee9a629f47df2ced6010851da27592c34dd)
- Update README with eval details (a17c3c86b7692201cb86b06bd0dbf9fce5b20b13)
- Update links to research blog (0d9ae7c0eb6d920a8cd892beaefbe3021354ff76)


Co-authored-by: Gang Li <[email protected]>

results/OrbyAgent-ActIO-72b/README.md

@@ -0,0 +1,7 @@
+### OrbyAgent-ActIO-72b
+
+This agent is developed by [Orby AI](https://www.orby.ai/).
+
+The agent does not use any benchmark-specific information in the prompts. For WebArena benchmark, we use the original evaluator and task definitions for fair comparison.
+
+It uses the ActIO model of 72B parameters as a backend, with both screenshot and HTML as inputs. More details can be found in our [research blog](https://www.orby.ai/resources/elevating-automation-orby-ais-generic-agent-framework-and-self-adaptive-interface-learning-technique).

results/OrbyAgent-ActIO-72b/miniwob.json

@@ -0,0 +1,16 @@
+[
+    {
+        "agent_name": "OrbyAgent-ActIO-72b",
+        "study_id": "orby-agent-v0-actio-v0-miniwob",
+        "benchmark": "MiniWoB",
+        "score": 64.2,
+        "std_err": 1.4,
+        "benchmark_specific": "No",
+        "benchmark_tuned": "No",
+        "followed_evaluation_protocol": "Yes",
+        "reproducible": "Yes",
+        "comments": "NA",
+        "original_or_reproduced": "Original",
+        "date_time": "2025-02-21 15:03:35"
+    }
+]

results/OrbyAgent-ActIO-72b/webarena.json

@@ -0,0 +1,16 @@
+[
+    {
+        "agent_name": "OrbyAgent-ActIO-72b",
+        "study_id": "b5fc5be7-54cc-4fc1-a9ee-73447b9c3eae",
+        "benchmark": "WebArena",
+        "score": 34.7,
+        "std_err": 0.25,
+        "benchmark_specific": "No",
+        "benchmark_tuned": "No",
+        "followed_evaluation_protocol": "Yes",
+        "reproducible": "Yes",
+        "comments": "Use original WebArena eval protocol and task definitions",
+        "original_or_reproduced": "Original",
+        "date_time": "2025-02-21 15:05:12"
+    }
+]

results/OrbyAgent-Claude-3.5-Sonnet/README.md

@@ -0,0 +1,7 @@
+### OrbyAgent-Claude-3.5-Sonnet
+
+This agent is developed by [Orby AI](https://www.orby.ai/).
+
+The agent does not use any benchmark-specific information in the prompts. For WebArena benchmark, we use the original evaluator and task definitions for fair comparison.
+
+It uses Claude-3.5-sonnet-20241022 as a backend, with both screenshot and HTML as inputs. More details can be found in our [research blog](https://www.orby.ai/resources/elevating-automation-orby-ais-generic-agent-framework-and-self-adaptive-interface-learning-technique).

results/OrbyAgent-Claude-3.5-Sonnet/miniwob.json

@@ -0,0 +1,16 @@
+[
+    {
+        "agent_name": "OrbyAgent-Claude-3.5-Sonnet",
+        "study_id": "orby-agent-v0-claude-3.5-miniwob",
+        "benchmark": "MiniWoB",
+        "score": 74.9,
+        "std_err": 1.2,
+        "benchmark_specific": "No",
+        "benchmark_tuned": "No",
+        "followed_evaluation_protocol": "Yes",
+        "reproducible": "Yes",
+        "comments": "NA",
+        "original_or_reproduced": "Original",
+        "date_time": "2025-02-21 14:54:16"
+    }
+]

results/OrbyAgent-Claude-3.5-Sonnet/webarena.json

@@ -0,0 +1,16 @@
+[
+    {
+        "agent_name": "OrbyAgent-Claude-3.5-Sonnet",
+        "study_id": "orby-agent-v0-claude-3.5-webarena",
+        "benchmark": "WebArena",
+        "score": 36.5,
+        "std_err": 0,
+        "benchmark_specific": "No",
+        "benchmark_tuned": "No",
+        "followed_evaluation_protocol": "Yes",
+        "reproducible": "Yes",
+        "comments": "Use original WebArena eval protocol and task definitions",
+        "original_or_reproduced": "Original",
+        "date_time": "2025-02-21 15:00:22"
+    }
+]