Georgia Institute of Technology
Motion Imitation and Adaptive Sim-to-Real Transfer for Robot Control
Abstract
dc:description.abstractDeep reinforcement learning (RL) combined with motion imitation has emerged as a powerful paradigm for learning control policies, enabling robots to acquire diverse motor skills by tracking reference motions. These methods, however, face two fundamental challenges: obtaining or generating high-quality and diverse reference motions at scale, and transferring policies trained in simulation to hardware, where dynamics mismatches cause failure. In this thesis, we present a suite of techniques that address these limitations by advancing the state-of-the-art in two key areas: scalable motion imitation and sim-to-real transfer. First, focusing on motion generation, we introduce a two-stage learning framework that eliminates the need for expensive motion capture data. Using RL with simplified dynamics, this approach automatically generates feasible motion plans, which are then tracked by an imitation policy that ensures real-world operability. Next, to scale this approach across hundreds of diverse behaviors and prevent catastrophic forgetting, we develop Adaptive Motion Sampling, which allows the policy to retain previously mastered skills while continuing to make progress towards learning new and challenging ones. For hardware deployment, we devised two complementary sim-to-real approaches. EnvMimic uses RL to model residual dynamics from a small number of real-world trajectories. BayRnTune expedites the transfer to hardware by intelligently adjusting simulation parameters to match reality, allowing the robot to build upon its existing knowledge rather than relearning for every adjustment. When applied to quadrupedal and buoyancy- assisted robots, our methods demonstrate a significant improvement in real-world performance while efficiently utilizing data. Beyond behavior, we broaden the scope of automation to embodiment itself: RobotDesignGPT leverages vision-language models to synthesize kinematically valid, simulation-ready robot designs from natural language descriptions and reference images, which integrate directly with our motion generation and imitation pipeline. Finally, we extend these techniques to whole-body humanoid control and investigate cross-simulator validation as a scalable proxy for hardware evaluation. By augmenting a large-scale humanoid motion imitation pipeline with automatic domain randomization driven by evaluation in a second, independent physics simulator, we show that sim-to- sim adaptation substantially improves robustness within the adaptation loop and flattens performance degradation under increasing perturbations. Our investigation further reveals that a single evaluation simulator is a useful but insufficient proxy for simulator-agnostic robustness. Together, this body of research establishes a framework for progressive refinement, spanning automated design synthesis, motion generation, scalable imitation, and data-efficient, validated real-world transfer.
Degree
thesis:*- Name thesis:degree_name
- Computer Science, PhD
- Department thesis:degree_department
- Computer Science
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Sontakke, Nitish
- Advisor dc:contributor.advisor
-
- Ha, Sehoon
- Committee members dc:contributor.committeemember
-
- Turk, Gregory
- Ravichandar, Harish
- Merel, Josh
- Kumar, Rajat
- ORCID person:identifier.orcid
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/82228