Ashwin Balakrishna

I am a Research Scientist at Google DeepMind working on building generally intelligent robots. Broadly speaking, I am excited about bridging the general-purpose capabilities of vision and language foundation models with algorithmic paradigms for learning from interaction, such as imitation learning and reinforcement learning. I've previously spent time at Toyota Research Institute and Nuro, where I worked on large-scale robot learning for robotic manipulation and autonomous vehicle planning respectively. I did my PhD in Computer Science in the AUTOLAB at UC Berkeley, where I developed a number of algorithms for safe and efficient online robot learning and studied applications to deformable manipulation, industrial automation, and robot grasping. I did my bachelors degree at Caltech in Electrical Engineering, where I worked on a number of applications of machine learning and signal processing to scientific problems.

Email  /  CV  /  Google Scholar  /  Twitter  /  LinkedIn  

Fully open-source vision-language-action model for general-purpose robotic manipulation. Exhibits strong performance both in the zero-shot and finetuning regimes.

Develops a safe reinforcement learning system for autonomous motion selection as part of the NuroDriver. Tested extensively in simulation and real world autonomous trials!

A large-scale cross-embodiment dataset and results suggesting positive policy transfer across robot embodiments.

A large-scale, in-the-wild robot manipulation dataset of 75K+ trajectories collected in various settings such as homes, labs, offices and more across multiple continents! Initial experiments suggest that co-training policies with DROID significantly improves policy robustness and OOD generalization.

A thorough investigation of what design choices matter most for building performant visually-conditioned language models. We release optimized and hackable training code, evaluation code, and all trained models for the community to build on.

A simple reinforcement learning algorithm that can be applied to any existing actor critic method to accelerate exploration for sparse reward tasks.

An algorithm for robust off policy comparison of learned reward functions.

Safe exploration by meta-learning risk measures across environments with different dynamics.

An algorithm for query-efficient interactive imitation learning which learns to cede control to a supervisor when (1) in novel states or (2) in bottlenecks where task success is unlikely.

Safe and efficient RL from image observations by leveraging suboptimal demonstrations to structure exploration and examples of constraint violations to satisfy user-specified constraints.

An algorithm for reducing supervisor burden by limiting context switches in interactive imitation learning.

A scalable and robust RL algorithm which optimizes for a combination of expected performance and tail risk under a distribution over learned reward functions.

An algorithm for safe reinforcement learning which utilizes a set of offline data to learn about constraints before policy learning and a pair of policies which seperate the often conflicting objectives of task directed exploration and constraint satisfaction to learn contact rich and visuomotor control tasks.

An MPC-based algorithm for robotic control (ABC-LMPC) with (1) performance and safety guarantees for stochastic nonlinear systems and (2) the ability to continuously explore the environment and expand the controller domain.

A new algorithm for safe and efficient reinforcement learning (SAVED) which leverages a small set of suboptimal demonstrations and prior task successes to structure exploration. SAVED also provides a mechanism for handling state-space constraints by leveraging probabilistic estimates of system dynamics.

A new formulation of imitiation learning from a non-stationary supervisor, associated theoretical analysis, and a practical algorithm to apply this formulation to develeop an RL algorithm which combines the sample efficiency of model-based RL and the fast policy evaluation enabled by model-free policies.

An algorithm for rapidly exploring large sets of grasps on objects with challenging geometries.

Learning to disentangle multiple cables in a variety of different knotted configurations.

Learning to kit novel 3D objects in complex 3D cavities with self-supervised rotation estimation.

Learning robust untangling policies with learned progress detection and recovery behaviors.

A general method for multi-task fabric manipulation using learned visual correspondences which can be applied across different robots to manipulate fabrics of varying shapes and colors.

An asymptotically optimal algorithm for rapidly learning to grasp new objects through online exploration.

Algorithms for untying dense knots in cables in the real world.

A new fabric simulator for learning fabric smoothing policies and learned policies which successfully smooth fabric in simulation and transfer to physical robotic systems.

A new algorithm for learning symmetry aware visual representations for deformable objects.

An algorithm for leveraging priors from general purpose grasping systems to accelerate online grasp exploration on novel, difficult to grasp objects.

A self-supervised algorithm which learns to orient unseen objects with unknown geometry given only a depth image observation of the desired orientation.

A method for multi task fabric manipulation by leveraging recent advances in video prediction and depth sensing.

An algorithm for learning visual correspondences for highly deformable objects and an associated controller which is used to manipulate rope into a variety of different arrangements either by learning from demonstrations or by designing interpretable geometric policies on top of the learned visual representation.

An efficient geometric algorithm (ClusterPush) for singulating a set of clustered planar objects.

Formulation and algorithms for the problem of efficiently identifying and retrieving a specific object in a cluttered environment.

High resolution measurement of both intraocular pressure and ocular pulsation profiles using an implmantable micro-optical sensor and portable optical detector.

Algorithms for rapid and reliable earthquake detection for earthquake early warning systems.

Algorithms for predicting electrical vehicle power usage for different charging network designs.

An evaluation of different machine learning algorithms for intraocular pressure measurement.

A new microscale implantable sensor and associated algorithms for accurate and convenient measurement of intraocular pressure.

A new wearable, wireless sensor to aid postoperative recovery from retinal detachment surgery.

A simple, wearable electromechanical sensor to aid postoperative recovery from retinal detachment surgery.

A neural network based algorithm to efficiently extract accurate intraocular pressure measurements from reflection spectra from an optics-based intraocular pressure sensor.

Evaluation of a new optics-based intraocular pressure sensor in live rabbits.

A piezoelectric based device to harvest power from human vocal cord vibrations.

Mathematical model and algorithm for controlling continuously flying solar aircraft.