Sim2Real Edge Actuation DRL Benchmark: 50 Trials on Model Compression

Parv Mittal's dataset from Harvard Dataverse contains 50 controlled empirical trials benchmarking Deep Reinforcement Learning policy compression. It evaluates trade-offs between model precision, matrix truncation, storage size, microcontroller latency, and energy efficiency for trajectory tracking. The dataset was last updated on June 13, 2026.

Use Cases

• Benchmarking the accuracy-latency trade-off of 8-bit quantization for DRL policies based on the described precision comparison.
• Evaluating the impact of Singular Value Decomposition truncation ranks on model performance and storage footprint.
• Estimating real-time operational energy efficiency for ARM Cortex-M based robotic actuators based on the described simulation latencies.
• Comparing binary storage footprints of compressed DRL policies for edge deployment.

Strengths

• Dataset comprises 50 highly controlled empirical trials, providing a structured benchmark.
• Explicitly evaluates multiple, specific trade-offs: 32-bit vs. 8-bit precision, SVD truncation ranks, storage, latency, and energy efficiency.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file size are unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific simulation environment and trials conducted.

Provenance

Source

Harvard Dataverse, authored by Parv Mittal.

Collection Method

Likely generated from experiments in a high-fidelity PyBullet simulation environment.

Freshness

Last updated 2026-06-13 14:30:46; freshness should be verified.